Retirement patterns change. When that happens, measures based on single-period cross-sectional data may provide a misleading picture of cohort retirement behaviour. We offer two cohort-based measures. One draws on income tax records to follow actual cohorts of individuals over time, the other on a time-series of cross-sectional surveys. We find that the retirement patterns based on the two approaches are similar but differ, often substantially, from single-period patterns. While pseudo-cohort measures can be assessed more quickly and at lower cost, knowledge of differences across income groups, income replacement rates, and so on, must rely on full longitudinal records.