Abstract: This paper is centered on model selection and model averaging procedure in varying coefficient partially linear models when the responses are missing at random. Under the misspecification framework, the focused information criterion (FIC) and the frequentist model average (FMA) estimator were developed based on the imputation method and the Profile least-squares technique. Then, theoretical properties of the FIC and FMA were examined. The simulation studies demonstrate the superiority of the proposed method and the approach will be applied to CD4 data.