When analyzing stock market data, it is common to encounter observations that differ from the overall pattern. It is known as the problem of robustness. Presence of outlying observations in different data sets may strongly influence the result of classical (mean and standard deviation based) analysis methods or models based on this data. The problem of outliers can be handled by using robust estimators, therefore making aberrations less influential or ignoring them completely. An example of applying such procedures for outlier elimination in stock trading system optimization process is presented.