An analysis of longevity in dairy cattle on a lactation basis is proposed. The approach allowed each lactation to have its own baseline hazard function, which gives a better description of the hazard than traditional analyses of the whole length of life. As a consequence, the overall fit of the model to the data was improved and fewer time-dependent variables were needed. Longevity on a lactation basis was defined from one calving to the next instead of from the first calving to culling. However, no new information was added and it was still the overall risk of being culled that was modelled. It is shown that no cow effect is needed in the lactation basis model because a censored record is not complete, a cow can appear as uncensored only once, and a cow cannot be censored after having been culled. Different subdivisions of the stage of lactation effect were tested and the first ten days of lactation were shown to correspond to an increased risk of being culled. There were no major differences in sire variance between the longevity analysed on a lactation basis and longevity based on the entire length of life.