Cereal producers are under pressure to increase yields and maintain profitability against a background of environmental constraints and high fertiliser costs. The production of high yields requires high inputs of N, and excessive N can lead to pollution of watercourses. This provides an incentive for the maximisation of nitrogen use efficiency (NUE), defined as grain yield per unit available soil N from all sources. Routes to the improvement of NUE may be through selection of an appropriate environment for the crop, better management or crop genetic improvement. However, the relative importance of these choices is poorly understood. Here we have used a modelling approach to quantify the effects of these factors on NUE. We performed an analysis using the Sirius wheat simulation model for a range of N treatments at two contrasting European sites: Rothamsted. UK and Seville. Spain. Several simple crop traits were selected for sensitivity analysis of NUE. These included traits controlling wheat development, determining sizes of N storage pools in the plant and traits responsible for uptake-efficiency of roots for water and N. We used Sirius because it is based on simple, mechanistic descriptions of wheat phenology and nitrogen uptake and redistribution, which makes it possible to link model cultivar parameters with simple physiological traits. Our analysis showed that weather and N management are the source of large variations in NUE. At Rothamsted, where water was not a limiting factor, N treatments produce more variation in NUE (similar to 51 %) than weather (similar to 32%). At Seville, where water is limited, weather was responsible for larger variation in NUE (for a shallow soil and low N treatment up to similar to 100%) compared with similar to 40% for N treatments. Two traits (leaf [N] and phyllochron) out of six showed potential for improvement of NUE. A decrease in leaf [N] increased NUE by 10-15%, when N was limiting, but for high N supply the effect on NUE was negligible. Increasing phyllochron to delay flowering produced up to 15% increase in NUE at Rothamsted, but no increase at Seville. Our analysis demonstrated that a crop simulation model is a powerful tool for deconvoluting complex traits in wheat. This may facilitate genetic and subsequent genonucs research by focusing experiments only on those wheat traits that are identified by the modelling study as the most promising.