Summary1. We had two objectives: (i) to determine the generality of, and extend the applicability of, a previously reported empirical relationship between leaf-level net photosynthetic rate ( A M , nmol g − 1 s − 1 ), specific leaf area (SLA, m 2 kg) and leaf nitrogen mass fraction ( N M , mmol g − 1 ); and (ii) to compare these empirical results with a mechanistic model of photosynthesis in order to provide a mechanistic justification for the empirical pattern. 2. Our results were based on both literature and original data. There were a total of 160 and 87 data points for the leaf-level and whole-plant data, respectively. 3. Our multiple regression for single leaves was ln( A M ) = 0·66 + 0·71 ln(SLA) + 0·79 ln( N M ), r 2 = − 0·80; only the intercept (0·11) differed for the whole-plant data. These results are not significantly different from previously published relationships. 4. We then converted the mechanistic model of Evans and Poorter, and a modified version which includes leaf lamina thickness ( T ) and leaf dry matter (tissue) concentration ( C M ), into directed acyclic graphs. We then derived reduced graphs that involved only T , C M , SLA, N M and A M . These were tested using structural equation modelling, with measured lamina thickness ( T ′ ) and leaf dry matter ratio (LDMR, g dry mass g − 1 fresh mass) as indicators of T and C M . The original Evans-Poorter model was rejected, but the modified version fitted the structural relationships well. The same qualitative models also applied to the whole-plant data, although the path coefficients sometimes differed. 5. Using simulations, we show that the original Evans-Poorter model predicts a positive correlation between SLA and N M that maximizes A M . The data closely follow this predicted relationship. The correlation between the actual values of A M (standardized units) and the predicted values obtained from the modified Evans-Poorter model was 0·74 and increased to 0·82 once three outlier points were removed. 6. These results provide a mechanistic explanation for the empirical trends relating leaf form and carbon fixation, and predict that SLA and leaf N must be quantitatively co-ordinated to maximize C fixation.