The objective of this study was to compare two methods for analysis of longevity in dairy cattle. The first method, currently used for routine genetic evaluation in the UK, uses a linear model to analyse lifespan, i.e. the number of lactations a cow has survived or is expected to survive. The second method was based on the concept of proportional hazard, i.e. modelling the conditional survival probability of a cow as a function of time. Comparisons were based on estimated heritabilities, ranking of estimated breeding values of sires, estimated effects of covariates used in the final models, and the distribution of residuals. The same data set, 21497 observations on the number of lactations cows had survived, was used for both analyses, even in the presence of censored observations. Cows in the data were progeny of 487 sires. Heritability estimates for lifespan or survival were approximately 0·06 for both methods, using the definition of heritability on a logarithmic scale for the proportional hazards model. Correlations between breeding values for sires were high, with absolute values ranging from 0·93 to 0·98, depending on the model fitted. It was concluded that it may be justified to use the standard Weibull model even for discrete time measures such as the number of completed lactations, but that more research is needed in the area of discrete time variates.