The hypothesis that the minimum error in predicted refraction after implantation of an intraocular lens (IOL) of calculated power is the sum of the random error in (1) the measurement of the axial length, (2) the measurement of the corneal power, and (3) the estimation of the pseudophakic anterior chamber depth (ACD) is proposed. Based on preoperative and postoperative biometry of 584 IOL implantations, 54% of the error was attributed to axial length errors, 8% to corneal power errors, and 38% to errors in the estimation of the postoperative ACD, when a fixed ACD was used in the IOL calculations. However, if the ACD was predicted according to a previously described regression method, the contribution of error from the ACD source was reduced to 22%, thereby reducing the total refractive prediction error from +/- 1.03 diopters (D) (+/- SD) to +/- 0.92 D (+/- SD). These predictions accord with clinical results.