Five latent heat flux (LHF) products are evaluated based on in situ observations in the South China Sea (SCS), including the ECWMF ERA‐Interim (ERA‐I), the NCEP2, the Objectively Analyzed air‐sea Fluxes (OAFlux), the Japanese 55 year Reanalysis (JRA55), and the TropFlux data sets. The results show that there are good correlations between the LHF products and observations, ranging from 0.68 to 0.74. However, mean biases of −8 to 40 W m−2 exist in the LHF products with respect to the observations. For root‐mean‐square errors, the OAFlux data set is the closest to the observations, followed by ERA‐I and TropFlux, while the NCEP2 data set shows significant overestimation. It is found that the biases in the near‐surface‐specific humidity are most correlated with the biases in the LHF products, followed by the biases in the near‐surface wind speed, air temperature, and sea surface temperature. The biases in the LHF products have a prominent seasonal variation that is 25 W m−2 higher in boreal winter than in summer. Using the thermal equation, it is shown that the tendency errors of the mixed‐layer temperature estimated by the biases in the LHF products vary from −2.0 to 3.5°C/month in the SCS. When all of the products are averaged, the errors are reduced to a range of −0.7 to 1.5°C/month. It is noteworthy that the errors in summer are more obvious than those in winter, since a thinner mixed layer in the summer can amplify the effect of even a small bias in the LHF.