Unbiased stereological methods were used in conjunction with microcomputed tomographic (micro-CT) scans of human and animal bone to investigate errors created when the parallel plate model was used to calculate morphometric parameters. Bone samples were obtained from the human proximal tibia, canine distal femur, rat tail, and pig spine and scanned in a micro-CT scanner. Trabecular thickness, trabecular spacing, and trabecular number were calculated using the parallel plate model. Direct thickness, and spacing and connectivity density were calculated using unbiased three-dimensional methods. Both thickness and spacing calculated using the plate model were well correlated to the direct three-dimensional measures (r(2) = 0. 77-0.92). The correlation between trabecular number and connectivity density varied greatly (r(2) = 0.41-0.94). Whereas trabecular thickness was consistently underestimated using the plate model, trabecular spacing was underestimated at low volume fractions and overestimated at high volume fractions. Use of the plate model resulted in a volume-dependent bias in measures of thickness and spacing (p < 0.001). This was a result of the fact that samples of low volume fraction were much more "rod-like" than those of the higher volume fraction. Our findings indicate that the plate model provides biased results, especially when populations with different volume fractions are compared. Therefore, we recommend direct thickness measures when three-dimensional data sets are available.