1. The rostral ventromedial medulla (RVM) participates in the modulation of nociceptive transmission by spinal cord neurons. Previous anatomic studies have demonstrated that RVM neurons project to laminae I, II, and V of the dorsal horn; laminae VII and VIII of the intermediate and ventral horns; the intermediolateral column; and lamina X. The RVM contains at least three physiologically defined classes of neurons, two of which, the ON and the OFF cells, have been implicated in nociceptive modulation. Because these cells classes are intermingled in the RVM, it has not been possible to determine the spinal laminar projection targets of ON and OFF cells by anatomic methods. Therefore in the current study we employed antidromic microstimulation methods to determine the laminar projections of two of the three classes of RVM neurons, the ON and the OFF cells. 2. In lightly anesthetized (with methohexital sodium) rats, single-unit extracellular recordings were made from 48 RVM neurons that were physiologically characterized as ON (30) or OFF (18) cells. The recording locations of 45 of these neurons were recovered. Thirty-seven were found in the nucleus raphe magnus and eight were located near its dorsal and lateral borders. 3. Thirty-two physiologically identified RVM neurons (18 ON and 14 OFF cells) were antidromically activated from the cervical spinal cord using a monopolar stimulating electrode. The stimulating electrode was moved systematically in the white matter until antidromic activation could be produced with currents of < or = 20 microA (6.1 +/- 0.7 microA, mean +/- SE). The points from which minimum currents were required to antidromically activate the neurons were located mainly in the ipsilateral dorsolateral funiculus (DLF) (27 of 32). In a few cases, lowest antidromic threshold currents were found near the border between the DLF and ventrolateral funiculus (VLF) or, rarely, in the VLF itself. In these cases, the cell recordings were found to be near the dorsal boundary of the RVM. 4. While one electrode was used to stimulate the parent axon in the lateral funiculus, a second was used to explore the gray matter for the presence of collateral branches. The identification of a branch was initially determined by an increase in antidromic latency. At the same rostrocaudal plane of the spinal cord, stimulation of the DLF induced an antidromic spike that invaded the neuron earlier than the antidromic spike elicited by stimulation in the gray matter. Collateral branches were confirmed by establishing that the location of the minimum threshold point for antidromic activation of the neurons from the second electrode was in the gray matter, that the minimum current required to antidromically activate the neuron from that point was too low to activate the parent axon in the DLF, and that a collision occurred between the spikes induced by the two stimulating electrodes. 5. In 17 cases, physiologically identified RVM neurons (10 ON and 7 OFF cells) were antidromically activated from the gray matter of the cervical spinal cord using a current of 8.4 +/- 2.1 (SE) microA. Minimum threshold points for antidromic activation were found in laminae I-II (3 ON and 4 OFF cells), lamina V (5 ON and 6 OFF cells), and regions ventral to the lateral reticulated area (3 ON and 2 OFF cells) of the gray matter. As indicated by these numbers, some neurons were antidromically activated from more than one gray matter region. In general, all OFF cells and 9 of 10 ON cells were antidromically activated from low threshold points in either laminae I-II or lamina V. 6. In six cases, neurons were activated from separate points located in two or three different laminae of the gray matter. Three OFF cells were activated from laminae I-II and V, one OFF cell and one ON cell were activated from lamina V and from more ventral points, and one ON cell was activated from laminae I-II and from points ventral to lamina V.