Armstrong's (1983) criticisms of PhiIip and Watson's (1982b) note on optimal drilling patterns are based on misinterpretation, supported by errors in geometry. Furthermore, Armstrong argues conceptually from the perspective of Matheronian geostatistics, as does also Rudenno (1983a, b). Thus, both authors embrace a global, probabiIistic method of assessing spatial data sets, an approach that must be contrasted strongly with local, deterministic interpolation methods. Some of the important assumptions of geostatistics are outlined and the non-unique nature of the method discussed. A probabilistic approach can be judged appropriate when exploration data are erratic, or known to be unreliable, or when extrapolation is desired. Where data are treated as exact, kriging can give the most biased result if tested against other interpolation methods. We reiterate our previous conclusion that the equilateral triangular grid is approximately 30 per cent more efficient (in terms of numbers of drill holes) compared with a square grid for the same maximum interpolation distance. This is simply due to the improved isotropy of sampling. New, locally-based, deterministic methods for computing ore reserves seem to be required.