Long-term total phosphorus loads from 17 urban catchments in the USA were predicted using five different measures of central tendency defining site mean concentration (arithmetic mean, geometric mean, median, harmonic mean and flow-weighted mean). Overall, the flow-weighted mean concentration produced the most accurate predictions of long-term loads. The geometric mean produced the second most accurate predictions. Along with the median and harmonic mean, the geometric mean predicted long-term load relatively well at most catchments exhibiting negative correlations between event mean concentration and total event runoff depth. However, they significantly underestimated long-term load at catchments exhibiting a positive correlation between these variables. Better estimates of long-term load at these two catchments were produced using the flow-weighted mean and arithmetic mean. However, the arithmetic mean tended to overestimate long-term load at the remainder of catchments.