Crop simulation models are increasingly being used to simulate the response of crop production to variation in input use. Current and widely used crop models differ strongly in the way in which green area index (GAI) and radiation use efficiency (RUE) is affected by nitrogen (N) supply. Three different methods of simulating effect of N on development of GAI were tested in combination with three different methods of simulating effects of N on RUE. The methods tested represent functions applied in three existing wheat simulation models: FASSET, Sirius and DAISY. GAI depends in FASSET on crop dry weight, temperature and N uptake, in Sirius on temperature and N uptake, and in DAISY GAI depends on dry weight and temperature. Sirius has no effect of N on RUE, DAISY uses a segmented linear response function, and FASSET uses a curvilinear response. The different methods were implemented in the FASSET model framework, and maximum RUE at optimal N supply was calibrated for each model combination using 4 years of growth analysis data from an experiment in winter wheat with three rates of mineral N fertiliser at Research Centre Foulum, Denmark. The model combinations were validated using 2 years of growth analysis data from an experiment at Research Centre Foulum with different timing of N application. The model combinations were tested against grain yield response to increasing N supply from a series of N fertiliser experiments in Denmark.The observed development of GAI and dry weight over time in the calibration and validation data sets could be reproduced by all combinations of GAI and RUE models. This shows that a large variation in N supply rates is more important than detailed sampling over time when validating and testing crop response to N supply. The observed response of grain yield to increasing rates of mineral N fertiliser could be reproduced by most of the model combinations. However, the yield increase was overestimated with the use of a segmented linear response of RUE to N supply, and the optimal N rate was underestimated when the N response of RUE was ignored.