In　this　study,　a　model　selection　procedure　for　varying-coefficient　model　based　on　longitudinal　data　is　proposed　to　distinguish　three　types　of　variables:　variables　not　in　the　model,　variables　in　the　model　with　time-independent　coefficients　and　variables　in　the　model　with　time-varying　coefficients.　To　identify　these　three　kinds　of　variables　simultaneously,　we　extend　the　present　variable　selection　method　from　cross-sectional　data　to　longitudinal　data.　This　method　combines　the　B-spline　function　approximation　and　Adaptive-Lasso　penalty　to　perform　variable　selection　and　do　nonparametric　estimation　simultaneously.　Validity　is　illustrated　with　a　set　of　simulation　experiments,　and　results　indicate　the　proposed　variable　selection　procedure　performs　well　in　distinguishing　the　real　type　of　independent　variables.