1. The effects of intracellular Mg2+ (Mgi2+) on the single N-methyl-D-aspartate (NMDA)-activated channel burst duration and frequency and on the mean NMDA-activated patch current were studied in outside-out patches from cultured rat cortical neurons. The inhibition by Mgi2+ of mean patch and whole-cell currents were compared, and some possible explanations for the observed differences were investigated. 2. The burst duration at +60 mV did not depend on Mgi2+ concentration, suggesting that the channel can close when blocked by Mgi2+. The number of bursts per second increased significantly in the presence of Mgi2+, suggesting that the rate of channel opening is higher when Mg2+ from the intracellular solution occupies its binding site. 3. Mgi2+ caused a voltage- and concentration-dependent inhibition of mean patch current. The inhibition is in quantitative agreement with the effects of Mgi2+ on the single-channel current and on burst parameters. 4. Based on the effects of Mgi2+ on burst parameters and on single-channel current, a four-state model in which the NMDA-activated channel can close while blocked by Mgi2+ is proposed. By fitting the model to the mean patch current data, we estimate that the rate of channel opening is increased by a factor of 1.4 when Mgi2+ occupies the channel. This estimation provides evidence that occupancy of the NMDA-activated channel by Mgi2+ destabilizes the closed state. 5. Mgi2+ reduced NMDA-activated whole-cell currents in a voltage- and concentration-dependent manner. However, normalized whole-cell and mean patch currents at positive voltages differed in two significant respects. First, when currents were recorded in a 0 Mg2+ pipette solution, whole-cell currents at positive voltages were smaller. Second, Mgi2+ appeared to inhibit whole-cell current less effectively than it inhibited mean patch current. 6. Inclusion of the Mg2+ chelators EDTA and ATP in 0 Mg2+ pipette solutions did not increase the whole-cell current measured at +60 mV. This observation suggests that the difference between normalized whole-cell and mean patch currents with 0 Mg2+ pipette solution was not due to block of whole-cell currents by residual Mgi2+. 7. When a pipette solution containing EGTA and Mg2+ was used to buffer Mgi2+, inhibition by Mgi2+ of the whole-cell current was enhanced, suggesting that the free Mg2+ concentration inside a neuron can remain below the pipette Mg2+ concentration. However, we cannot exclude other explanations for the differences between the inhibition by Mg2+ of mean patch and whole-cell currents.