Ion cyclotron resonance frequency, f, is conventionally converted to ion mass-to-charge ratio, m/z (mass "calibration") by fitting experimental data spanning the entire detected m/z range to the relation, m/z = A/f + B/f(2), to yield rms mass error as low as ~200 ppb for ~10,000 resolved components of a petroleum crude oil. Analysis of residual error versus m/z and peak abundance reveals that systematic errors limit mass accuracy and thus the confidence in elemental composition assignments. Here, we present a calibration procedure in which the spectrum is divided into dozens of adjoining segments, and a separate calibration is applied to each, thereby eliminating systematic error with respect to m/z. Further, incorporation of a third term in the calibration equation that is proportional to the magnitude of each detected peak minimizes systematic error with respect to ion abundance. Finally, absorption-mode data analysis increases mass measurement accuracy only after minimization of systematic errors. We are able to increase the number of assigned peaks by as much as 25%, while reducing the rms mass error by as much as 3-fold, for significantly improved confidence in elemental composition assignment.