A multi-dimensional database includes data organized into hierarchical levels. For each level within a dimension, predefined thresholds are set to determine which intersections are precomputed and which are not. The general result is to precompute intersections which take relatively longer to compute at run time, and to not precompute intersections which can be calculated relatively quickly at run time. When multiple dimensions of data are provided, a step-by-step process uses the level threshold for each dimension to determine whether data is to be prestored at an intersection. An odd-even calculation insures that computations performed at run time are relatively efficient. The technique results in good run time response, while not requiring maximum storage or recalculation time for data updates. By adjusting the thresholds at the individual dimension level, response of the multi-dimensional database can be changed as desired.