Lake model provides boundary conditions for numerical weather forecasting model in terms of sensible heat, water vapor and momentum flux. Because the mechanism controlling water vapor and heat fluxes at different temporal scales are different, it is necessary to make multiple time-scale offline evaluation for lake models. In this study, the CLM4-LISSS model (Community Land Model version 4-Lake, Ice, Snow and Sediment Simulator) was driven by the meteorological and radiation measurement data at the Bifenggang site of the Taihu Eddy Flux Network from year 2012 to 2016, and the simulation results were compared with the measurement of eddy covariance system. The light extinction and roughness length parameterizations were optimized in pursuit of best simulation on annual mean latent heat flux. The lake surface temperature, water vapor and heat fluxes from half-hourly to annual time scales were used to evaluate model performance. The results indicated that simulated lake surface temperature agreed well with observation for all time scales, but the diurnal range was underestimated. The model was able to capture the half-hourly and annual variation trends of latent heat flux, but obvious biases occurred for monthly values with overestimation in fall and winter and underestimation in spring and summer. Parameterization of light extinction coefficient may attribute to the biases of lake surface temperature and latent heat flux. The agreement between simulated and observed sensible heat flux was not as good as that of latent heat flux, even though the trend of annual mean sensible heat flux can be captured by the model. The CLM4-LISSS model can capture the variability of latent and sensible heat flux under cold front condition, but obviously overestimated the latter one during strong wind period. This paper can provide valuable information to the application and development of lake models.