The accuracy of two simple methods was compared for the prediction of crude protein (CP) content of above-ground plant material of mixed-species composition on abandoned cultivated land in Japan. The first method is based on standard CP values (in g kg−1 dry matter) for individual species (STV method) as listed in the literature. The second procedure (GLM method) was an application of the generalized linear model using the relative above-ground biomass of monocots and legumes, total herbage mass, and day of year. Predictions were made at the quadrat scale, and for surveyed sites based on average of values for five or six quadrats in a single survey. A ‘leave-site-out’ method was adopted for model validation of the generalized linear model. The observed values of CP content ranged between 21·5 and 161·9 g kg−1 dry matter (DM). With the STV method, the values of root mean square error (RMSE indicates average estimation error) were 50·9 at the quadrat level and 53·8 at the surveyed-site level (both g kg−1 DM). When a ‘leave-site-out’ validation was carried out, the RMSE-values for the GLM method were 23·2 at the quadrat level and 13·2 at the surveyed-site level (both g kg−1 DM). We therefore propose adoption of the GLM method for the purpose of estimating the CP content in herbage on abandoned sites.