The present study determines numerical densities (NVsyn) and total numbers of synaptic discs in laminae I-IV of the rat S2 dorsal horn. Previous methods for NVsyn have the advantage of being relatively simple, but these assume that the discs are round, flat, and of uniform size. In our material, serial reconstructions indicate that these assumptions are not met. Accordingly we use a stereological method that is not as dependent on these assumptions. This method is to divide the surface density of the discs by the mean surface area of a disc (NVsyn = SVsyn/Ssyn). We refer to this as a reconstruction method because synaptic discs are reconstructed from serial sections. We also calculate numerical densities by several previously used standard methods, and the findings are similar but not identical. We find that numerical density and total synaptic numbers are smallest in lamina I, and densities and total numbers are not significantly different when lamina II is compared to laminae III and IV. Thus the intense labeling of terminals with certain compounds that characterize lamina I and II does not imply an increase in total synaptic numbers or in synaptic density. In addition there is a general increase in synaptic densities and numbers as one proceeds from lamina I to lamina IV. Another point is that the numerical density of synapses in the dorsal horn is approximately that of the cerebral cortex. These data will serve as a basis from which to judge the effects of denervations and other manipulations that purportedly change synaptic numbers.