This paper explores sociolinguistic variation in the frequency of apologising in the spoken part of the British National Corpus. The starting point for the investigation is the �apology formula�, represented by the lexemes �afraid�, �apologise�, �apology�, �excuse�, �forgive�, �pardon�, �regret� and �sorry�. The sub-corpus used for the study comprises a spoken text mass of about five million words and represents dialogue produced by more than 1700 speakers acting in a number of different conversational settings. More than 3000 examples of apologising form the basis for the analysis. In the BNC, young and middle-class speakers favoured the use of the apology form. Only minor gender differences in apologising were apparent. The study implies that formulaic politeness is an important linguistic marker of social class and also shows that corpus linguistic methodology can successfully be used in socio-pragmatic research.