The analysis of positron emission tomography (PET) images at the pixel level may yield unreliable parameter estimates due to the low signal-to-noise ratio of pixel time activity curves (TAC). To address this issue it can be helpful to use techniques developed in the pharmacokinetic/pharmacodynamic area and referred to as 'population approaches.' In this paper, we describe a new estimation algorithm, the Global-Two-Stage (GTS), and assess its performances through Monte Carlo simulations. GTS was compared to the basis function method on synthetic [11C](R)-PK11195 data, and to weighted nonlinear least squares on synthetic [11C]WAY100,635 data. In both cases, GTS produced parameter estimates with lower root mean square error and lower bias than the well-established estimation methods used for comparison, with a negligible increase of computational time. GTS was applied first to all the pixels of the simulated slices. Then, after a preliminary segmentation of pixels into more homogeneous populations, GTS was applied to each subpopulation separately: this last approach provided the best results. In conclusion, GTS is a powerful and fast technique that can be applied to improve parametric maps, as long as preliminary estimates of parameters and of their covariance are available.