Multilevel models are used extensively in social and behavioral science research because the models are able to accept hierarchical data structures. However, when multicollinearity among fixed effects of the model exists, multicollinearity may lead to imprecise coefficient estimates. We investigate a new method of estimating fixed effect coefficients in multilevel model when multicollinearity exists. The proposed method of estimating parameters is based on ridge regression. We apply this method to student assessment data and compare the results with an existing method. The proposed method provides coefficient estimates which have smaller variance than the existing method. Furthermore, we present PRESS statistic which is adapted to the proposed method. Results suggest that the proposed method predicts data better than the existing method.