A 0.5 g sample is far too small. I have calculated (by using Pierre Gy's sampling theory the relative standard deviation of the Fundamental Sampling Error (FSE) for some concentrations and the minimum sample size (Ms) if 10 % standard deviation is acceptable. I have assumed that 95 % of the gold nuggets pass a 0.5 mm sieve.

Au content Ms = 0.5 g RSD = 10 %
g/ton min RSD (%) min Ms (g)

1000 36 6.4
100 110 64
10 360 640
1 1100 6400

FSE is the minimum theoretical sampling error for a given material which assumes that the analyte particles are randomly distributed in the matrix material. If there clustering or segregation (as there nearly always is) the actual sampling error may be much larger.

In order to succeed with your analysis you have to extract the gold into the analytical sample from a much larger sample than 0.5 g.

How were the previous results generated? Sample size is a good place to start, but may not be the only cause. What variability did your field and lab duplicates show? What did you standards show?

Larger sample sizes with fire assay or BLEG methods could help.
Something that was not asked by the previous posters was how much has the field sampling changed between the two surveys? Was the time of year different? How consistent were the two sampling methods? Did both surveys really collect equivalent material?

Answering these will be as important as sorting out a good prep and digestion method for the samples.

Be careful here. Using Gy's method to determine an appropriate sample size, if that is your problem can lead to errors (specifically, an under-estimation of the necessary sample size).

Gy's sampling theory is based on binomial statistics, which is appropriate for the sampling of major or minor elements. However, because gold typically occurs as rare nuggets, binomial statistics are not appropriate; Poisson statistics are. I am constantly surprised how many people do not realize Gy's method is a 'binomial method', and so try to use Gy's method for gold. Most of the published efforts applying Gy's method for gold sampling problems have been relatively unsuccessful, and I guess there is no surprise why.

The best reference for applying Poisson statistics to this sampling problem is in Clifton et al. (1969 - USGS Professional Paper). I would start there, as that will identify the correct sample size for your problem.

From previous experience with analysing sediments in gold exploration I found 5 grams of sample sufficient to detect the gold at ppb to ppm levels. This would have been extracted in organic solvent after dissolution and analysed with Graphite furnace AAS! Just be very careful of the grind size and the resultant nugget effect due to wrong particle size.

Thanks for the brilliant insight. I'll check that out and see how it goes.

Totally agree. There are many variables with stream sediments sampling. BLEG is the best to use, but keep in mind screening and sampling should be constant.

0.5 g and 5 grams are ridiculously small sample size for gold in general and for stream sediments specifically. BLEG sample usually is 1 kg with 0.5 mm maximum size. It allows to give and accurate realistic figures on ppb level. Accuracy of 10% with ICPMS is deceiving. It is instrumental accuracy, but not the one of the whole analysis, which is prone to sampling errors. The last can easily be 10000 % due to poor sampling /small sample mass.

5 g sample is not even enough for Au assay after pulverizing 1 kg and then reducing it to 5 g unless you deal with rock without nugget effect.

There is a basic old empirical Richards-Chechott equation that was used to calculate the minimum necessary mass of a sample in exploration and mineral processing in USSR and Russia since 1930s:

m (kg)= kd2 , where m – minimum sample mass in kg, k – coefficient of homogeneity (varies usually from 0.05 to 1.0), d – maximum diameter of particles in millimetres. K depends on distribution of the component in the sample. For gold in stream sediments k is usually 1-1.5. For placers k can go up to 3.6.

You may look up also later publications by Kreiter V.M., 1968 Geological Prospecting and Exploration Mir Publisher, Moscow 384 p. and also Handbook on mineral processing Editor: O.S. Bogdanov, V.I. Revnivtsev, Moscow, Vol. 1-4, 1982-1984.

Apologies for my large contribution, but was thinking about some old sediment data recently.

In exploration, IMO, too much is often read into absolute values returned from stream sediment samples. There are too many sampling factors to back-calculate the true grade of the sediment from an assay result of a sub-sample of a certain size fraction. We should not be looking for a representative sample of the sediment in the stream (unless we are placer miners) but should look to answer the question "does the stream contain gold? Yes or No!"

However, in the case at the start of this discussion, the small assay sample size is probably the main issue.
Sample preparation for sediments (excluding BLEG) usually takes the fines after sieving at 80mesh (180microns) so you are going to lose all gold grains larger than this - the only way to capture larger grains is by using multiple screen assays or using a BLEG method (I must admit to not liking BLEGS mainly due to the logistics of collecting and transporting large samples in remote areas).

If your gold passes the 180micron sieve (it usually does) the effect of having gold grains at e.g. 100 microns can still be very scary. By making some assumptions you can calculate how many grains you get in a certain size fraction of sediment at a known grade. An old rule of thumb for sampling is that to get repeatable results you need 20 particles of gold in your assay sub-sample. Assuming a scenario of:

Grains that are 1/3 of a sphere in shape (sphere diameter = d)
d = 100 microns
Grade of material = 0.25ppm
sub-sample = 50g

You will get only 3 to 4 gold grains in each 50g sub-sample!

Now comes the frightening part; you have a sample with 75g of such material and the lab splits your fine fraction to create one final 50g assay sub-sample - you have a high chance of getting only 2, 1 or possibly 0 grains of gold in your assay sub-sample if the lab does not do a very good splitting job. This may happen if the assayer assumes the sub-sample is homogeneous and scoops the assay sub-sample from the top of the pulp envelope - gold grains, may have settled to the lower part of the envelope during transport.

Anyhow, the best practice I have found to minimise these issues in a field program is:

Know your samplers – collecting all samples for the day from a pile of sand outside the pub is not unknown!
Know the grain size and shape of gold in streams; use a screen assay on coarser fractions or a larger BLEG sample if you have a coarse gold problem. Coarse gold in this sense is around 50 microns or larger.
Use as large an assay sub-sample as the lab allows, most of the major labs offer an aqua regia digest on a 30 or 50g aliquot (e.g. ALS method Au-TL44, 50g sub-sample digested by aqua regia followed by ICP-MS to detect down to 1ppb).
Collect a -2mm sample in the field that when sieved by the lab provides slightly #less# fine material than the ideal assay sub-sample weight. This prevents you from having to split the fine-fraction due to having too much material. If you end up with more fine-fraction than the lab can digest in one sub-sample then assay each of the sub-samples and use the highest returned gold value when assessing the data.
Use a reputable lab and insert sufficient QAQC check-samples to monitor each batch of samples. Potential problem = if a batch fails QAQC you have no material to get re-assayed. So take a field duplicate at every sample site and store the duplicates until the assays on the primary samples have passed QAQC.