Noninvasive recording of arterial input functions using regions of interest (ROIs) in the left ventricular (LV) chamber obviates the need for arterial cannulation in PET, but it is compromised by the limited recovery coefficient of the LV chamber and by statistical noise. In the present study, a new mathematical model has been developed, which corrects for the spillover of radioactivity both from the myocardium into the LV ROI and the blood into the myocardial ROI. The method requires the measurement of a time-activity curve in the LV chamber during the dynamic H2(15)O PET study and the measurement of the recovery coefficient of the LV ROI using a 15O-carbon monoxide (C15O) scan and venous blood sampling. This approach was successfully validated against direct measurements of the arterial input function using an on-line beta detector in five greyhounds undergoing dynamic H2(15)O PET imaging. This technique also yielded myocardial blood flow (MBF) values which were not significantly different from those obtained with the beta-probe analyses (maximum difference less than 2%), provided that the LV ROIs were sufficiently large to provide good counting statistics. When this model was not applied for large ROIs (small recovery in LV ROI), systematic overestimations in MBF compared with beta-probe analysis (e.g., a factor by 40% for a recovery coefficient of 0.7) were observed. Thus, this technique enabled the prediction of an accurate input function using the LV time-activity curve, and hence, noninvasive quantification of MBF without arterial cannulation.