The purpose of this study was to assess the reproducibility of repeated positron emission tomography (PET) measures of brain monoamine oxidase B (MAO B) using deuterium-substituted [11C]L-deprenyl ([11C]L-deprenyl-D2) in normal subjects and to validate the method used for estimating the kinetic constants from the irreversible 3-compartment model applied to the tracer binding. Five normal healthy subjects (age range 23-73 years) each received two PET scans with [11C]L-deprenyl-D2. The time interval between scans was 7-27 days. Time-activity data from eight regions of interest and an arterial plasma input function was used to calculate lambda k3, a model term proportional to MAO B, and K1, the plasma to brain transfer constant that is related to blood flow. Linear (LIN) and nonlinear least-squares (NLLSQ) estimation methods were used to calculate the optimum model constants. A comparison of time-activity curves for scan 1 and scan 2 showed that the percent of change for peak uptake varied from -18.5 to 15.0% and that increases and decreases in uptake on scan 2 were associated with increases and decreases in the value of the arterial input of the tracer. Calculation of lambda k3 showed a difference between scan 1 and scan 2 in the global value ranging between -6.97 and 4.5% (average -2.1 +/- 4.7%). The average percent change for eight brain regions for the five subjects was -2.84 +/- 7.07%. Values of lambda k3 for scan 1 and scan 2 were highly correlated (r2 = 0.98; p < 0.0001; slope 0.955). Similarly, values of K1 showed a significant correlation between scan 1 and scan 2 (r2 = 0.61; p < 0.0001; slope 0.638) though the values for scan 2 were generally lower than those of scan 1. There was essentially no difference between the values of model constants calculated using the NLLSQ or LIN methods. Regional brain uptake of [11C]L-deprenyl-D2 varied between scan 1 and scan 2, driven by the differences in arterial tracer input. Application of a 3-compartment model to regional time-activity data and arterial input function yielded lambda k3 values for scan 1 and scan 2 with an average difference of -2.84 +/- 7.07%. Linear regression applied to values of lambda k3 from the LIN and NLLSQ methods validated the use of the linear method for calculating lambda k3.