Prediction of aggregate gradation from 2-D images of aggregates typically involves two stages—estimation of aggregate gradation by number and then by weight. For the first part, one needs to determine whether a given aggregate would pass through a given sieve size or not, and repeat this for all the aggregates under consideration. For the second part, one needs to estimate the volume of each of the aggregates to obtain their weights—since aggregate gradation is generally expressed in terms of weight proportion. Both these stages accumulate errors during computation because of the attempt to predict 3-D features from the acquired 2-D data. To reduce these errors, the present paper suggests incorporation of two shape-dependent correction factors for these two stages. Tests are conducted on a set of aggregates to obtain these correction factors. Thereafter, the proposed scheme is implemented on a fresh set of aggregates for validation.