It is known that Logistic Regression coupled with Partial Least Squares dimension reduction (PLSDR-LD) is capable of extracting a great deal of useful information for classification from gene expression profile and getting a rather high classification accuracy rate. In this study, we replace the logistic function of Logistic Regression with several functions which are similar to logistic function in appearance, and apply these functions to the analysis of microarray data sets from two cancer gene expression studies. We compare these newly introduced models with PLSDR-LD proposed in the literature. The most effective models with good prediction precision are lastly provided through analyzing the results of two experiments.