A detailed nonlinear model, the 3s-Beta model, for photothermal responses of flowering in rice (Oryza sativa L.) was evaluated for predicting rice flowering date in field conditions. This model was compared with other three models: a three-plane linear model and two nonlinear models, viz, the modified rice clock model (m-RCM) and a logistic model. Two existing multilocational data sets for photoperiod sensitive and nearly insensitive genotypes were used to evaluate the models. For a photoperiod-sensitive cultivar, nonlinear models described the data more accurately than the linear one; the performance of the three nonlinear models was similar although the logistic model overpredicted days to flowering in a tropical environment. For nearly photoperiod-insensitive genotypes, the models were evaluated first using experiments started in 1984. The models performed best in the order of the 3s-Beta, m-RCM, logistic, and linear. When the coefficients derived from the 1984 experiments were used to predict flowering dates observed in experiments of 1983, relative performance of the models remained the same although the differences became smaller. The results related to model performance and methods for model parameterization under field conditions were discussed.