The RAS method extrapolates a single matrix to conform to new row and column totals. This paper proposes a cell-correction of RAS (CRAS) that uses the deviations of multiple RAS projections, to improve the projection of the input-output table (IOT) of a specific country or region. The new method is tested on eleven survey-based IOTs of Spanish regions for 1999-2005. CRAS is shown to outperform RAS when three to four survey IOTs are used that are close to the target IOT. When more IOTs are added, for most but not all regions, CRAS gradually becomes worse than applying RAS to the single best IOT.