Various retention models have been developed and adopted to predict the retention behaviors of solutes in high performance liquid chromatography. Although the retention times can be successfully predicted in a gradient elution, it is difficult to predict the fundamental parameters of the eluted peaks, such as the number of theoretical plates, the resolution between neighboring peaks, as well as asymmetry factors. Thus far, the Snyder retention model has widely been used to predict the retention behavior in both isocratic and gradient conditions. However, it has a critical defect in that it cannot closely follow the retention behaviors of solutes when a relatively low content of organic modifier is used in the mobile phase. This is a result of the nonlinear relationship between the logarithm value of the retention factor and the volumetric percentage of organic modifier in the mobile phase. To overcome this shortcoming, a modified retention model was adopted to predict the retention times in several linear gradient conditions. A numerical method that transforms any gradient condition into discrete step gradient conditions was also proposed to predict the retention times in a gradient elution. The model is suitable to apply to nonlinear and multilinear gradient conditions, including actual obtained gradient profiles. Two kinds of organic modifiers, methanol and acetonitrile, were employed, and four purine compounds were used as solutes. The predicted retention times obtained by the modified retention model are in good agreement with experimental data in terms of the achieved gradient conditions.