Year × variety means of wheat grain crude protein content (CP) and Hagberg falling number (HFN) surveyed in the UK from 1974 to 1993 were investigated with a residual maximum likelihood (REML) analysis using variety, nitrogen fertilizer application (kg N/ha), average monthly temperatures (°C), and total monthly rainfall (mm) as fixed terms, and year as a random term. Annual variety means were weighted according to the inverse of their variances. The statistical significance of fixed terms as they were added to the model was evaluated by comparing the Wald test statistic with critical values of chi-squared. CP, except for 1976 which was associated with prolonged drought conditions, was best described with a model combining effects for variety (P < 0·001), nitrogen application rate (0·0143, S.E. = 0·00312, P < 0·001), July temperature (0·184, S.E. = 0·0582, P < 0·01) and May rainfall (−0·00952, S.E. = 0·00233, P < 0·001). A model describing HFN combined effects for variety (P < 0·001), August rainfall (−0·930, S.E. = 0·2579, P < 0·001) and June temperature (24·4, S.E. = 7·37, P < 0·001). These models were evaluated for describing CP and HFN of cv. Avalon grown on sandy-loam from 1982 to 1992 at one site. Correlation coefficients of 0·51 (P = 0·1) and 0·72 (P = 0·012) were obtained between observed values at this site and those predicted from the nationally derived models for CP and HFN respectively. Variety effects suggested that the varieties commonly used in the mid 1970s would give higher CP and lower HFN than their more modern counterparts if grown at similar levels of nitrogen and in the same climatic conditions.