Structural equations modelling (SEM) procedures are widely used today in quantitative research on HRM, for testing complex models of causality, incorporating several latent variables. The usual procedure of estimation of such models is based on analysis of covariance relations between latent variables, implemented in widely used software (Lisrel, Amos, EQS…). The purpose of this article is to present and discuss an alternative method of SEM estimation based on analysis of variance: the Partial Least Square (PLS) analysis. We will see that this method is suitable for the test of complex causal models, typical of situations met in the domain of HRM. In the first section, we summarized essential traits of the PLS approach. The second section presents an empirical application, via the estimation of a research model of contingent workers commitment. This study was conducted on a sample of 208 temporary workers. We can establish an indirect impact of temporary contracts characteristics on organizational commitment of the temporary workers. This impact was mediated by subjective work insecurity.