Comparisons of the regression equation between two methods for testing rocks and ores at the Dukat gold-silver deposit are made using two algorithms. In the first loss function defined by the sum of squares of residuals (deviations from the values of the dependent variable of the regression line), in the second-the sum of the absolute values of the residual normalized to sum values of the dependent and independent variables. As a result, it was shown that in the first case, in order to increase the accuracy of estimates of the coefficients, it is necessary to exclude observations in which the absolute value of the residuals normalized to their standard deviation exceeds the critical level of 5-7. In the second case, it is stable to sample formation and the number of observations can be reduced by 2-3 times. Under these conditions, the error in the estimates of the coefficients turns out to be close.