Experimental aging research is very dependent on the determination of the survival characteristics of the animal species or strain under study. Such data are generally inferred from mortality curves of cohorts of animals that are set aside at an early age for aging studies. Rectangular survival curves and the presence of multiple pathological lesions are a prerequisite for aging studies so as to resemble the situation in man. From 1977 onwards, many rat cohorts have been formed in the Institute for Experimental Gerontology (IVEG) for the study of aging processes. Data from these have been analysed for a period of 5 years up to and including 1982. (Males and females of the WAG/Rij and BN/BiRij strains were used.) The 50% survival and the maximum survival of cohorts varied considerably, but showed no consistent trend over the years. The median (50%) survival between the cohorts differed by as much as 7.9-10.7 months for the strains and sexes studied. Maximum survival between the cohorts varied from 3.7 to 9.9 months. Median and maximal survival were greater for the females. Maximum survival and 50% survival correlated significantly, the relation between the two being approximately linear. The effect of removing animals from cohorts on the estimation of 50% survival was only minor, whereas maximum survival was clearly diminished by this procedure. The wide variation in survival characteristics, even between successive cohorts, cautions against too simple a measure of the animals survival in only one number for median or maximal survival in months. An indication of the variance of 50% survival and of maximum survival should therefore be included in scientific publications. Moreover, the 50% survival is the parameter of choice to define cohorts, not only because this can be most reliably estimated with good confidence limits, but also because this measure is the least sensitive to removing animals from the cohorts. As this will often be the case in many research institutions, it might be of practical importance to order old animals from different cohorts since this diminishes the chance of using an extremely short or long lived cohort. Finally, the analysis revealed that combining intact or incomplete cohorts into larger survival curves resulted in nearly identical graphs. An attempt was made to calculate the minimum cohort size which yields survival curves with constant 95% confidence limits.