A habitat suitability index (HSI) model, developed for the American oyster,Crassostrea virginica, along the Gulf of Mexico, was field tested on 38 0.1-ha reef and nonreef sites in Galveston Bay, Texas. The HSI depends upon six (HSI1) or, optionally, eight (HSI2) variables. The six variables are percent of bottom covered with suitable cultch (V₁), mean summer water salinity (V₂), mean abundance of living oysters (V₃) (a gregarious settling factor), historic mean water salinity (V₄), frequency of killing floods (V₅), and substrate firmness (V₆). The optional variables are the abundance of the southern oyster drillThais haemostoma (V₇), and the intensity of the oyster pathogenPerkinsus marinus (V₈). The HSI values were lowest at high and low salinity sites and highest at intermediate-salinity sites. To validate the model, the hypothesis that the output of the HSI model was correlated with oyster density was therefore tested. A significant correlation was found between HSI1 and oyster density (Kendall Tau Beta correlation coefficient, τ=0.674, p<0.001, n=38); however, a statistical independence problem exists with the above test, that is, oyster density is both the independent standard for the test and a variable in the model. A regression model was constructed to test the relationship between log-transformed oyster density values (dependent variable) and the other variables of the model (independent variables). Most variation (r²=0.72, r=0.85) in the log-transformed density values were explained by a regression model that contained V₂, V₄, V₅, V₆, V₇, and V₈ as independent variables. The regression model was useful in constructing a modified HSI model (MHSI). A significant correlation (τ=0.674, p<0.05, n=10) was found between MHSI1 values and oyster densities from reefs closed to harvesting. The MHSI improves upon the original model by (i) simplifying the model structure, (ii) removing the requirement to measure V₃, (iii) accounting better for the negative effects of high salinity, disease, and parasitism upon oysters, and (iv) eliminating the statistical independence problem by dropping V₃ from the model. The MHSI should be tested against a new, independently-collected data set.