The aim of this study was to evaluate the suitability of Illius and Gordon’s simulation model for predicting the digestibility and intake of tropical roughages by ruminants. Data from seven suitable empirical studies (i.e. studies that reported all requisite data needed to run the model) in which tropical roughages were fed to ruminants were used. These seven studies had 235 intake and 70 digestibility measurements on goats, sheep and cattle. The coefficient of variation (cv) of the observed digestibility was 13.8%. A linear least square regression relationship between the observed (Y) and predicted (X) digestibility accounted for barely 37% of the variation for the entire data set [Y = 0.24(s.e. = 0.056) + 0.61(SE = 0.100) X] and for 53% of the variation when 5% of the observations were treated as outliers:[Y = 0.17 (SE = 0.047) + 0.72 (SE = 0.084) X]. For both equations the intercept and the slope were different from zero and unity, respectively. Although, a plot of residual digestibility showed no distinct pattern, this model should be used with extreme caution because of its inherent noise. The observed intake had a higher cv of 18.3%. Regression relationships between the observed and predicted intakes accounted for barely 15% of the variation for the entire data set: [Y = 30.1(SE = 7.39) + 0.71(SE = 0.111) X] and for 20% of the variation when 5% of the extreme values were treated as outliers: [Y = 28.7(SE = 6.73) + 0.75(SE = 0.101) X]. A residual plot showed that the model systematically underestimated the intake of roughages which elicited high intakes. The discussion was structured to identify, justify and propose pathways for enhancing the model.