Depth-distance dependent bias in body-wave magnitude is investigated using the International Seismological Centre (ISC) dataset from 1978 to 1993. Large deviations in mb determination at varying epicentral distances using the Gutenberg and Richter (1956) depth-distance correction terms shows that the correction terms presently used need to be revised. New empirical global depth-distance correction terms B(Δ,h) for body-wave magnitude are determined using the values of scalar moment M0 in the Harvard Centroid Moment Tensor (CMT) catalog to calibrate P-wave amplitude-distance curves. Comparison of event magnitude mb calculated using the Gutenberg-Richter correction terms, and that of this study, shows that the new depth-distance correction terms tend to increase small magnitudes and decrease large magnitudes. Comparison of the correction terms of Gutenberg and Richter, Veith and Clawson (1972), and Lilwall (1987b) with those obtained in this study shows that mb is biased most when using the correction terms of Gutenberg and Richter (1956). A systematic bias in the estimated mb for distance greater than 88° is observed for the Veith-Clawson and the Lilwall correction terms. Using Gutenberg-Richter, Veith-Clawson, and Lilwall correction terms, mb for deep earthquakes is systematically underestimated by about 0.1-0.15, 0.3-0.5, and 0.1-0.2 magnitude units, respectively. Application of the new correction terms to the ISC dataset shows that estimated mb is then independent of distance and focal depth and hence provides unbiased estimates of mb in comparison with other correction terms. This observation is also a positive test of the hypothesis that calibration with respect to M0 gives a better estimate of mb. Comparison of the standard deviation of mb values for single events using different depth-distance correction terms shows that the Gutenberg-Richter standard deviations are, on average, larger than those of Veith and Clawson, Lilwall, and this study. Application of station corrections obtained from magnitude residuals reduces the average standard deviations by about 0.07 magnitude unit, which is statistically significant.