INTRODUCTION AND OBJECTIVES: Several models based on ordinary differential equations have been proposed in the literature for cells proliferation. In order to verify which model better represent in vitro experiments performed with DU-145 human prostate cancer cell line, the Approximate Bayesian Computation (ABC) algorithm of TONI et al. (2009) was applied for model selection and parameters calibration. This algorithm becomes very attractive when the likelihood of the data is not available in an analytical way or when it is computationally very expensive (TONI et al. (2009), Costa et al. (2017)). Instead of using the likelihood function, the ABC algorithm is based on a set of successive populations evolving with more restrictive tolerances. MATERIAL AND METHODS: Four mathematical models were analyzed: Logistic Model, Gompertz Model, Richards Model and Generalized Logistic Model. The information about the DU-145 prostate cancer cell growth was measured each 24 hours during seven days. The Euclidian distance between the experimental observations and the responses of the mathematical models was used as the selection criterion based on the preview stipulated tolerances. RESULTS AND CONCLUSION: The ABC method with 2000 particles was applied for model selection and estimation of cells proliferation parameters. Model parameters were considered with uniform priors and uniform transition kernels. In order to solve the inverse problem, the four mathematical models were solved using Runge-Kutta 4th order. The vector of tolerances for the sequential populations of ABC code started in 5.4187e+05 and finished in 0.0542 e+05, covering a total of fifty-seven populations, where the last tolerance was imposed in accordance to the assumed measurement uncertainty (1%) and to the Morozov’s discrepancy principle. The Richards Model and the Generalized Logistic Model were selected providing accurate results to the number of cells varying with the time.