Corpus linguists are good at thinking about how best to gather, record, and annotate language samples. They have sometimes seemed less successful at exploiting corpora to generate findings of serious human interest, beyond the narrow fields of dictionary-making and language-teaching. However, results that have begun to emerge from the British National Corpus, if borne out by larger-scale research, are potentially significant for social policy and for our understanding of human nature. One section of the BNC samples the conversational speech of a cross-section of the British population. Having equipped a small subset of this material with grammatical annotations, I looked for correlations between speakers’ demographic properties (sex, age, region, social class) and the grammatical complexity of their utterances. Famously, in the 1960s Basil Bernstein claimed that speech complexity correlated with social class, but the BNC data do not bear this out. On the other hand there is a clear, statistically-significant correlation with age. Speech appears to grow more complex not just between childhood and maturity (which is not surprising) but throughout adult life (which is). The social implications of this finding are sufficiently weighty that it merits testing against larger data-sets and more languages.