The granule size distributions of starches isolated from eight bread wheat flours were determined. Breads were baked from the same flours by varying fermentation time, mixing speed and work input according to a central composite experimental design. Response surface equations were calculated for the bread characteristics loaf volume, weight, form ratio (height/width) and crumb firmness. Multivariate regression (PLS2) was used to relate the starch granule size distribution spectra to the coefficient matrices. The models obtained by PLS2 explained 55% and 48% of the total variance in the response surface coefficients for bread weight, and form ratio respectively, with correlation coefficients 0·74 and 0·69. No models were obtained for crumb firmness and loaf volume. The correlations obtained with the constant terms indicated that high weights and form ratios were promoted by small A-granules (size 12 μm). The effect of starch granule size distribution on the bread characteristics was modified by work input and mixing speed, while fermentation time had no effect.