The hypoxic zone in the northern Gulf of Mexico varies spatially (area, location) and temporally (onset, duration) on multiple scales. Exposure of fish to hypoxic dissolved oxygen (DO) concentrations ( 2 mg L.sup.-1) is often lethal and avoided, while exposure to 2 to 4 mg L.sup.-1 occurs readily and often causes the sublethal effects of decreased growth and fecundity for individuals of many species. We simulated the movement of individual fish within a high-resolution 3-D coupled hydrodynamic water quality model (FVCOM-WASP) configured for the northern Gulf of Mexico to examine how spatial variability in DO concentrations would affect fish exposure to hypoxic and sublethal DO concentrations. Eight static snapshots (spatial maps) of DO were selected from a 10 d FVCOM-WASP simulation that showed a range of spatial variation (degree of clumpiness) in sublethal DO for when total sublethal area was moderate (four maps) and for when total sublethal area was high (four maps). An additional case of allowing DO to vary in time (dynamic DO) was also included. All simulations were for 10 d and were performed for 2-D (bottom layer only) and 3-D (allows for vertical movement of fish) sets of maps. Fish movement was simulated every 15 min with each individual switching among three algorithms: tactical avoidance when exposure to hypoxic DO was imminent, strategic avoidance when exposure had occurred in the recent past, and default (independent of DO) when avoidance was not invoked. Cumulative exposure of individuals to hypoxia was higher under the high sublethal area snapshots compared to the moderate sublethal area snapshots but spatial variability in sublethal concentrations had little effect on hypoxia exposure. In contrast, relatively high exposures to sublethal DO concentrations occurred in all simulations. Spatial variability in sublethal DO had opposite effects on sublethal exposure between moderate and high sublethal area maps: the percentage of fish exposed to 2-3 mg L.sup.-1 decreased with increasing variability for high sublethal area but increased for moderate sublethal area. There was also a wide range of exposures among individuals within each simulation. These results suggest that averaging DO concentrations over spatial cells and time steps can result in underestimation of sublethal effects. Our methods and results can inform how movement is simulated in larger models that are critical for assessing how management actions to reduce nutrient loadings will affect fish populations.