This paper explores two different methods of tracing a specific speech act in a historical corpus. As an example, the development of apologies is investigated in the two hundred years covered by the Corpus of Historical American English (COHA, 1810-2009). One method retrieves apologies through their typical illocutionary force indicating devices (IFIDs), such as sorry, excuse, apologise and pardon, while the other retrieves passages in which apologies are explicitly mentioned (metapragmatic expression analysis). Both methods require a considerable amount of manual analysis of retrieved hits, which has to be verified through elaborate inter-rater reliability testing. The searches are restricted to fictional texts because they show a greater frequency of apologies than the alternative genres available in COHA, and they often allow the identification of behaviour as apologetic because it is discursively described as such by the fictional characters or the narratorial voice. The results show that the frequency of apologies increased considerably throughout the period covered by COHA. In the earliest period the IFID sorry was no more frequent than pardon and forgive. In the most recent period its frequency has multiplied almost six-fold and is more than three times larger than all the others taken together. The metapragmatic expression analysis allows an analysis of the development of strategies used to perform apologies. IFIDs have become more important while Taking on Responsibility and Explanation receded somewhat in their frequencies. On the basis of these results it is speculated that the force of apologies has decreased. What used to be sincere requests for exoneration has in many cases turned to token displays of regret.