Two-point calibration is often used to correct for nonuniformities across focal plane arrays (FPAs), as well as for calibration. Because the input-output curves of FPA channels are nonlinear, two-point calibration produces a systematic calibration error as a function of flux, and the channel-to-channel variations of this calibration error leave a significant post-correction nonuniformity. A physical model of detector nonlinearity is used to illustrate these points. A simple formula is proposed, which fits the input-output curves much better than the straight line used by two-point calibration, and is almost as easy to use. When the new formula is used, the system's performance is no longer sensitive to the choice of calibration temperatures, and no longer degrades rapidly outside the calibration interval.