The detection of two atypical values in a linear model is considered. The case of two atypical values in the same direction is studied in detail. The first Bonferroni inequality is used to obtain higher nominal percentage points. For a random sample of a normal distribution, the proposed statistic is reduced to the well-known Murphy test statistic. Similarly, for two atypical values in either direction, our statistic is reduced to the statistical studentised range. We investigate an approximate relationship between the critical points of the Murphy test and the range statistics studied.