The paper deals with modeling of human-like reaching movements using probabilistic models. The main issue under study is the generation of asymmetric velocity profiles with features compatible with previously obtained experimental data. For this purpose a popular minimum variance model is exploited and analyzed. A reformulated version of this model that does not feature a post-movement time specification is proposed and tested under simulation. It is shown that under proper selection of weight coefficients human-like motion profiles can be generated in the reformulated version of the minimum variance model.