A P index was developed as a tool to rank agricultural fields on the basis of P loss vulnerability, helping to target remedial P management options within watersheds. We evaluated two approaches, a soil P threshold and components of a P index, by comparing site vulnerability estimates derived from these two approaches with measured runoff P losses in an agricultural watershed in Pennsylvania. Rainfall-surface runoff simulations (70 mm h(-1) for 30 min) were conducted on 57 sites representing the full range of soil P concentrations and management conditions found in the watershed. Each site was comprised of two, abutting 2-m2 runoff plots, serving as duplicate observations. For sites that had not received P additions for at least six months prior to the study, Mehlich-3 P concentration was strongly associated with dissolved P concentrations (r2 = 0.86) and losses (r2 = 0.83) in surface runoff, as well as with total P concentration (r2 = 0.80) and loss (r2 = 0.74). However, Mehlich-3 P alone was poorly correlated with runoff P from sites receiving manure within three weeks prior to rainfall. The P index effectively described 88 and 83% of the variability in dissolved P concentrations and losses from all sites in the watershed, and P index ratings exhibited strong associations with total P concentrations (r2 = 0.81) and losses (r2 = 0.79). When site-specific observations were extrapolated to all fields in the watershed, management recommendations derived from a P index approach were less restrictive than those derived from the soil P threshold approach, better reflecting the low P loads exported from the watershed.