APPENDIX H

CALCULATION OF ESTIMATION VARIANCE FOR ROCK CREEK BULK SAMPLE

Memo
To	Henrik Thalenhorst;John	File No.
	Odden;Doug
	Nicholson;Dominique Francois-
	Bongarcon
From	Harry Parker	Cc
Tel
Fax
Date	May 30 2004

Subject Calculation of Estimation Variance for Rock Creek Bulk Sample

1.0 INTRODUCTION AND SUMMARY

On May 17, 2004 Henrik Thalenhorst provided a list of blocks for a base-case bulk sample and an expanded-case bulk sample. The base-case bulk sample contains 16 blocks of dimension 5 X 10 X 5 m in X, Y, Z. There are eight blocks on the 80 m bench and eight blocks on the 75 m bench. These blocks are centered on 525 N and extend from 475 to 515 E. In an expanded case 5 blocks are added centered on 535 N on the 80 m bench and 5 blocks are added centered on 515 N on the 80 m bench. These blocks extend from 475 E to 500 E. The base case has 10,800 t and the expanded case has 17,550 t.

Henrik asked what the estimation accuracy would be if holes at various spacings were used. The holes will be drilled at a 45 degree angle to the east. The analysis assumes that the pattern would be repeated (though offset) for both core and RC holes.

Table 1 shows relative standard errors and relative 90 % confidence limits for four spacings:

A.	Six holes on 10 m spacing in X and Y, with Y alternating between 520 and 530 North.
B.	Eleven holes on 5 m spacing in X and 10 m spacing in Y alternating between 520 and 530 North.
C.	Thirty holes on 5 m spacing in X and Y, in three lines centered on 520, 525 and 530 North.
D.	Fifty holes on 5 m spacing in X and Y, in five lines centered on 510, 515, 520, 525, 530, 535 and 540 N.

AMEC E&C Services Inc.
980 Lincoln Avenue No. 200
San Rafael, California
USA 94901
Tel      +1 650 504 0229
Fax     +1 415 482 1001
www.amec.com

2

Table 1: Relative Standard Errors and 90 % Confidence Limits

Data Configuration	Base Case (10,800 t)		Expanded Case (17,550 t)
	Relative Standard Error of Estimate	± 90 % Relative Confidence Interval	Relative Standard Error of Estimate	± 90 % Relative Confidence Interval
Case A	32 %	53 %	31 %	51 %
Case B	25 %	41 %	23 %	38 %
Case C	15 %	24 %
Case D			12 %	19 %

It is planned to estimate the grade of the bulk sample from offset patterns of core and RC holes. To establish a correction factor for exploration data, the estimates must be accurate. Table 1 shows that Cases A (envisioned in the program design) and B are inaccurate. Cases C and D could be used to establish approximate adjustment factors for Core and RC data, but these still could be less accurate than desirable.

The base case comprises 4 SMUs and the expanded case comprises 6.5 SMUs. It is clear from this analysis that accurate segregation of ore and waste will be extremely difficult, even if sampling is carried out using a 5 m grid. All that can be hoped for is to delineate broad zones of ore and waste. This must be confirmed at the feasibility stage using conditional simulation.

2.0 SETUP

The variogram model used is that for Zones 1+2, Table 7-3, AMEC 2004 Resource Report. This is a double nested unit sill exponential model with C0 = 0.502, C1= 0.437 and C2=0.060. The practical ranges in the N-S direction are 16.3 and 76.3 m. In the W direction dipping 83 degrees the ranges are 17.1 and 553.0 m. In the E direction dipping 7 degrees the ranges are 10.0 and 680.3 m.

The squared coefficient of variation for 5 m composites is 2.14, drawn from Zone 1, Table 7-4, AMEC 2004 Resource Report. The relative ordinary kriging variance is:

Rokv = (ok variance, unit sill variogram model)(Coefficient of Variation)²

The relative standard error of estimate is Rokv^0.5 .

The ±90 % confidence interval is approximately ± 1.645 (Rokv^0.5 .)

The discretization spacing for ordinary kriging was 2.5 m in X and Y and 5 m in Z. In a special version of AMEC’s single-block kriger, the 5 X 10 X 5 m blocks are aggregated and discretized.

3

The attached spreadsheet setup.xls shows the configuration of holes and blocks. Hole “collars” represent pierce-points on the 85 m bench. The holes at 520, 525 and 530 m north have four composites extending from the 85 m elevation downward at an inclination of 45 degrees. This provides coverage for blocks on the 75 and 80 m benches and one composite underneath. No composites “above” the blocks are used, as these probably would come from the weathered zone and be suspect. For the expanded case, the holes on the 510, 515, 535 and 540 bench have two composites extending from the 85 m elevation, as the incremental bulk sampled blocks are all on the 80 m bench.

Strathcona Letterhead

June 9, 2004
Joe R. Piekenbrock	E-MAIL AND MAIL
NovaGold Resources Inc.	joep@novagold.net
7853 Red Fox Drive
Evergreen, Colorado 80439 USA

Dear Joe:

Rock Creek Bulk Sample

Following our lengthy discussions on a bulk sampling program for the Rock Creek project, we are proving a summary of our thoughts on the factors to be considered, and the analysis that has been done to date to guide a sampling program on this challenging gold deposit.

Background and Definition of Sampling Challenge

The Rock Creek project is an example of a low-grade gold deposit with relatively coarse gold, one of the most difficult situations when attempting to quantify mineral resources and reserves. After a considerable in-fill drilling campaign conducted by Alaska Gold Company in 2003, Amec have recently completed an estimate of the mineral resources for the project. One of their main observations was that different drilling techniques have given rise to different gold grade regimes which in turn result in significantly differing resource estimates for the project, depending on which type of drill results are being used. Given the low-grade nature of the deposit, the question of which drilling technique provides the most reliable data is of great importance. To solve the question, AMEC have concluded that bulk sampling is required, a conclusion that we share in principle.

Based on the analysis by Amec of 29 twin drill hole results, the difference in gold assay results is in the order of 25 to 30%, with reverse circulation (RC) drilling showing systematically higher values than core holes, even after the removal of low-recovery intervals from the core holes and the removal of obvious or suspected contamination problems in RC holes. To be effective, a bulk sample should have an expected sample error considerably smaller than the variance between the two types of drilling. As well, the grade

Page 1

estimation variance for the bulk sample itself should also be well below the drill hole grade bias.

Bulk Sample Error

We have developed a protocol for the bulk sample, that can be summarized as follows:

$

Field Activities. Crush the entire sample (nominally 10 000 tonnes) to 95% passing 1.27 cm (2 inch). Split out 160 tonnes (1.6%) using a sample tower. Crush 160 tonnes to 95% passing 0.64 cm (3 inch) using the rolls crusher on the sample tower. Split out 20 tonnes (an additional split of 12.5%) as the final field sample.

$

Further Sample Comminution, Splitting and Assaying. Ship the field sample to F. L. Smidth Inc. Process Laboratories in Bethlehem, PA for dry grinding to 95% passing 100 microns (140 mesh) in a vertical mill. Ship the pulverized sample to SGS Lakefield Research in Ontario for treatment in a Knelson concentrator to recover gold particles >30 to 40 microns. The gold content of the Knelson concentrate will be put into solution by intense cyanidation and then determined by solution assay. From the Knelson tailings a one-eighth sample (12.5%) will be split out with a pulp splitter and subjected to intense cyanidation. Assaying of the final tailings will complete the gold grade determination process.

The sample error for this protocol, assuming four sub-samples of 2500 tonnes each at gold grades of 0.5, 1.0, 1.5 and 2.0 g/t, respectively (a head grade of 1.25 g/t for the entire lot), is around ∀7% for the whole lot, at the 95% confidence level.

Grade Estimation Variance for the Bulk Sample Tonnage

In preparation for the bulk sample test, the bulk sample area needs to be drilled off with the competing drilling techniques in some detail, and a separate gold grade estimate would be prepared for each technique used. Two of these techniques would repeat, as exactly as possible, the two historical methods giving rise to the estimation problems for the entire deposit.

Harry Parker of Amec has prepared a memo dated May 30, 2004 in which he evaluates the grade estimation variance for the bulk sample for different tonnages and for different drilling densities. His conclusions are contained in a table, reproduced below:

Page 2

Table 1 - Relative Standard Errors and 90 % Confidence Limits

Data Configuration	Base Case (10 800 tones)		Expanded Case (17 550 tonnes)
	Relative Standard Error of Estimate	∀ 90 % Relative Confidence Interval	Relative Standard Error of Estimate	∀ 90 % Relative Confidence Interval
Case A	32 %	53 %	31 %	51 %
Case B	25 %	41 %	23 %	38 %
Case C	15 %	24 %
Case D			12 %	19 %

Case ASix holes [for each type of drilling] on ten-metre spacing in X (grid-east) and Y (grid-north), along the centre line.

Case BEleven holes [for each type of drilling] on five-metre spacing in X and on ten-metre spacing in Y along the centre line.

Case CThirty holes [for each type of drilling] on a five-metre spacing in X and Y, in three lines. And finally

Case D Fifty holes [for each type of drilling] on a five-metre spacing in X and Y in five lines.

We have not yet received an update of these calculations that would spell out the tonnage and drill pattern required for a ∀ 90% relative confidence interval of say ten percent that we think would be required for the bulk sample to provide the grade certainty needed. However, it is obvious from Table 1 that a substantial tonnage of perhaps 25 000 tonnes or more, and an extensive drilling program on five-metre centres would be required.

Discussion and Conclusions

The sample protocol chosen for the Rock Creek bulk sample would yield a sample error of ∀ 7% that would allow a decision to be made which of the grade data are the more realistic to use for the resource estimate of the Rock Creek deposit and for its economic evaluation. However, a large bulk sample with a very substantial pre-sample drill program on five-metre centres would be required to reduce the forecast grade variance for the bulk sample to a compatible level.

There are three possible outcomes to such a bulk sample exercise:

Page 3

1.	The grade forecast based on diamond drilling is verified. In this case, the project would be evaluated using the AUKR4 case, the risk-adjusted grades with the RC grades reduced mathematically to match what the DDH grades would have been, based on the twin-hole analysis.
2.	The grade forecast does not support either the previous core or RC results, but is somewhat intermediate. In this case, the more cautious approach of using the AUKR4 case would be retained.
3.	The grade forecast based on reverse circulation drilling is verified. In this case, the grade data as is, case AUKR5, would be used for the evaluation of the project. However, upgrading of the diamond drill core grades, currently constituting two- thirds of the project database, would not be possible without a significant program of additional RC drilling using best practices to improve the statistical base that compares RC and DDH gold grades.

With only a limited knowledge of the project itself, but after having spent a few hours looking at part of the drill core left from the 2002 and 2003 diamond drilling programs, we are of the opinion that there is a greater chance that a bulk sampling program would verify the diamond drill core grades, rather than showing the core sample grades to be biased low.

Our initial estimate for the cost of the bulk sample Case A, while not entirely firm, came to about $1.3 to $1.5 million. The much larger program required to achieve a reasonable estimation variance for the anticipated gold grade of the bulk sample would probably double that figure. There is also, as we have pointed out, a real problem in securing a sample tower in time for this summer=s program.

It is not obvious that the Rock Creek project would warrant such a large outlay for a program that, if the results are negative, would add nothing but certainty to a marginal grade estimate, and that, if positive, (a less likely outcome), might add 125 000 ounces to its mineral resource inventory (at a cut-off grade of 0.6 g/t) by moving below-cut-off material above that line, without truly lifting the project above the realm of marginality because of the virtually unchanged gold grade.

There is the alternative to extend the scope of the existing twin hole data by re-drilling all of the 29 sites in such a way that, at the end, each site has a full complement of one old-style and one best-practice RC hole, one Aregular@ diamond drill hole repeating the 2003 drilling protocol, and one large-diameter (PQ) triple-tube DDH. Sampling of the RC holes would include a significant number of field duplicates (say 20%). The PQ core would be sampled in toto, in one-half metre intervals. The sample preparation and assay protocols would follow the 2003 practice.

Page 4

One would hope that, for the 29 sites combined, there would be very little grade difference between the triple-tube DDH and the best-practice RC holes, thus establishing the true grade of the intersections which in turn would be used to make a call as to which of the currently existing grade forecasts are more believable. It is anticipated that such a program would require some 4000 metres of drilling and would thus be expected to be a more economically sensible approach.

We regret that the planning for a bulk sampling program for the Rock Creek project has raised some questions about the economics and benefits, and scale of program required to achieve reliable and useful results, but hope that our comments contribute to NovaGold arriving at the most appropriate decision as to what the next step should be.

Yours sincerely,

H. Thalenhorst

cc: Harry Parker (harry.parker@amec.com)

Page 5