Linguists often rely on synchronic generational differences in language to supply evidence of language change in progress in "apparent time," yet this approach must always be evaluated against the possibility that such differences reflect change over speakers' lifetimes ("age grading"), rather than language change. The present paper compares apparent-time data on Montreal English with "real-time" data from earlier studies of the same community, in order to test the assumptions of the apparent-time model. The comparison reveals that, while some age-correlated lexical variables show stability over speakers' lifetimes, clearly suggesting ongoing change, others show changes in progress combined with change over speakers' lifetimes. However, the nature of individual change is generally found to be not the rejection of new variants by older speakers associated with the age-grading model, but late adoption of new variants by adults who learned older variants as children. Most postacquisition change therefore accelerates rather than retards change in progress. Evidence from two phonological variables suggests that late adoption is most characteristic of lexical variation. In any case, the possibility of late adoption implies that an accurate view of language change only emerges when both apparent- and real-time data are examined.