The Optical Backscatter Sensor (OBS) has been widely used to measure suspended sediment concentration in both field and laboratory conditions, even though it is very sensitive to many factors. The most significant factor suggested is the grain size. In order to enhance the quality of OBS data, an improved approach is proposed, based on the "mixture of linear component response" method (Green and Boon III, 1993) to account for the effect of grain size. In addition to an original sediment sample, which commonly serves as a single calibration material, an accompanying sediment sample is necessary to calibrate OBS sensors. A multi-fraction sediment model is used to predict the grain size distribution in suspension. Compared with existing methods, the improved approach does not require a sieving procedure nor the assumption that the sediment fractions of the calibrated sediment sample exhibit the same sensitivities as those of the suspended sediments. The applicability of our method has been verified by a series of laboratory experiments over silt-sand mixtures. The results show that this method successfully yields continuous concentration profiles, which agree well with the measurements using a suction method. The converted concentrations of the time-averaged OBS measurements by traditional method and by the improved method, respectively, are compared with the suction measurements. The results of linear regression analyses show that the coefficient of determination increases (e.g. from 0.55 to 0.92 for wave-current conditions) and the Root Mean Square Error decreases (e.g. from 0.97 to 0.39 for wave-current conditions). It demonstrates that improved method enhances the quality of OBS conversion. Furthermore, suggestions on selecting the accompanying sediment sample (i.e. on grain size and composition) are given.