Previous studies have identified both linear and curvilinear relations between increasing bite-force magnitude and the integrated electromyogram (EMG) of jaw-closing muscles. In an attempt to explain the discrepancy, bite forces of incrementally increasing magnitude were produced on the right-hand side in five specified directions by eight humans. Linear regression lines were fitted to normalized EMG activities of the left and right masseter and temporalis muscles against increasing bite-force magnitude in each direction. The grand mean of linear correlation coefficients was 0.79 (+/- 0.11 SD), suggesting an overall linear relation. Each set of individual data was fitted with polynomial lines up to the third order. The best fit was selected by statistical significance of coefficients and the least-square analysis of the sum of residues for each fitted line; 62% of individual data-sets were best fitted with linear regression lines, 31% with quadratic lines and the remaining 7% with cubic lines. Repeated analysis of residue variance of the pooled data showed that either a linear or quadratic line fitted every data set except one, for which a cubic line had the best fit. Working-side muscles had significantly larger linear correlation coefficients than corresponding balancing-side muscles for most bite-force directions. Analysis of variance of linear correlation coefficients revealed that the degree of linearity often depended upon the roles played by a muscle in producing forces in different directions. It appears that linearity or non-linearity of the EMG force relation is a determinant, among other variables, of the direction of the resultant force.