PURPOSE: The prognostic value of factors used in clinicopathologic staging of localized soft tissue sarcoma (STS) of the extremity were analyzed comprehensively. PATIENTS AND METHODS: Four hundred twenty-three patients with STS that was confined to the extremity were admitted to Memorial Sloan-Kettering Cancer Center from 1968 to 1978. Cox models for the hazards rates of tumor mortality, development of a distant metastasis, strictly local recurrence, and postmetastasis survival were developed. Tests of changes in the prognostic value of the important variables over time were performed, as well as an analysis of the effect of a local recurrence on the hazard rate of distant metastasis. RESULTS: Three unfavorable characteristics contained independent prognostic value for the rates of distant metastasis and tumor mortality: high grade (P less than .00001), deep location (P less than .0002), and size greater than or equal to 5 cm (P less than .007). Their Cox model coefficients did not differ significantly (P greater than or equal to .65); thus, a staging scheme based on the risk of ever developing a distant metastasis would assign equal prognostic weights to grade, depth, and size. The tumor grade effect during the initial 18 months was much larger in magnitude than those for depth and size, and its effect disappeared beyond that time (P = .0003). Thus, a staging scheme based on the risk of early metastatic spread would assign a distinctly larger prognostic weight to grade and lesser but equal weights to depth and size. There was no local recurrence effect on the rate of distant metastasis in the high-risk group (high grade, deep, and greater than or equal to 5 cm; P = .75), but there was a significant association among the remaining groups combined (P = .0039). The magnitude of this association actually increased according to the number of favorable characteristics presented (P = .0024). CONCLUSIONS: The refinement of clinicopathologic staging may depend on the choice of outcome variable: ultimate prognosis versus early metastatic spread. Additionally, the observed local recurrence effect may be explained by a tendency for some patients to acquire one or more unfavorable risk factors at the time of local recurrence.