Criteria are developed for the maximisation of the amount of information available from an analytical calibration graph. For linear calibration graphs, the methods involve replicate determinations of standards at two points, which are chosen so that they occur at the extremes of the calibration range. The increase in precision that can be obtained by using an optimised design can be considerable, particularly when the variance in the assay errors is not constant. The methods described can be extended readily to non-linear calibration graphs.