Abstract: The objective of this study was to estimate the genetic parameters of productive life of Chinese Holsteins, and to compare the prediction reliability and stability of different evaluation models. The genetic parameters of productive life records of 90 049 Chinese Holstein cows from 29 herds in Beijing were estimated by single-trait animal model (STAM), single-trait sire model (STSM) and multiple-trait animal models (MTAM). All models used for genetic parameter estimation took into account the fixed effects of herd, birth year-season and group of age at first calving, the random effect of individual additive genetic effect (animal model) or sire genetic effect (sire model), as well as the residual effect. The heritability of productive life derived from the single-trait animal model and single-trait sire model were 0.052 and 0.047, respectively. Multiple-trait animal models were relatively stable, with estimated heritability between 0.057 and 0.069, and the genetic correlations among these traits ranged from 0.779 to 0.998. For the data with large amount of censored records, multiple-trait animal models based on the first 3 lactations are more appropriate, meanwhile these model are more stable. It is the first genetic evaluation for productive life of Chinese Holstein, and these results provide theoretical basis for improving economic efficiency of farms and facilitating balanced breeding in the future.