The nucleolar area of uveal melanomas, measured from standard hematoxylin-and-eosin-stained microslides, has been shown in previous work to correlate well with survival following enucleation of the tumor-bearing eye. Despite this correlation, the accuracy of the original system for measuring nucleolar area was affected by several sources of error: the algorithm by which area was computed underestimated true area; geometric and optical factors caused overestimation of nucleolar area unless measured in the center of the microscopic field of view and the area of small nucleoli, as the result of several possible mechanisms, contains less useful information than the area of large nucleoli. This paper presents methods introduced to reduce error from these sources and demonstrates that the predictive value of nucleolar measurements is relatively insensitive to high levels of random variation.