In the literatures, there are mainly two approaches to model latent classes with covariates: one-step and three-step approaches. In the one-step approach, unknown membership and related covariates are taken into account simultaneously in the latent class model; nevertheless, several disadvantages have been figured out. On the contrary, the estimation of three-step approach is typically partitioned in to three parts: building a latent class model, assigning subjects to latent classes, and predicting latent classes with interested covariates. However, as warned by Bolck, Croon, and Hagenaars (2004), the classification error in the second step may attenuate the relationships between latent classes and covariates. Moreover, Bolch et al. have proposed a new three-step approach to correct parameter estimates of covariates. Thus the aim to this study is to provide modified procedures to overcome several limitations of the original BCH approach.