The TRMM Large Scale Biosphere-Atmosphere (LBA) experiment, conducted between January and February of 1999 in Southwest Amazon, deployed among other instruments NASA's C-band Doppler radar (TOGA) and four dense rain gauge networks. This paper presents a procedure devised to derive surface rainfall rate estimates from combination of TOGA observations and the in situ rain gauge rainfall measurements. The spatial and temporal scales considered are 2 ¿ 2 km2 grids of instantaneous to hourly rain accumulations. The procedure includes evaluation of TOGA calibration through comparisons with TRMM Precipitation Radar (PR) data and implementation of an optimal quantitative precipitation estimation (QPE) algorithm. Comparisons with PR indicated a 4-dB calibration offset occurring in the later two thirds of the observation period. The implemented QPE algorithm applies a parameter that differentiates the Z-R conversion in convective and stratiform regimes and a stochastic filtering approach for estimation of mean-field bias on the basis of radar-rain gauge comparisons at the hourly timescale. The calibration of the algorithm parameter values is formulated as a global optimization problem, which is solved by minimizing the radar-rain gauge rainfall accumulation root-mean-square (rms) difference at the hourly timescale. A random resampling calibration/validation exercise is performed to evaluate the algorithm performance and its sensitivity to parameter values. Validation against gauges shows that the algorithm produces unbiased estimates with ~57% relative RMS difference at the hourly scale. Comparison with S-POL rain estimates showed good correlation (0.9) but some overestimation (9%). Rainfall products are used to derive rainfall statistics for two distinct meteorological low-level wind regimes (easterly and westerly) that occurred during LBA. Finally, instantaneous rain estimates are compared against TRMM PR rainfall profiles for six coincident storm cases showing high correlation (0.9) and low (7%) systematic difference (PR overestimation).