We tested metrics from full-waveform (FW) LiDAR (light detection and ranging) as predictors for forest basal area (BA) and aboveground biomass (AGB), in a tropical moist forest. Three levels of metrics are tested: (i) peak-level, based on each return echo; (ii) pulse-level, based on the whole return signal from each emitted pulse; and (iii) plot-level, simulating a large footprint LiDAR dataset. Several of the tested metrics have significant correlation, with two predictors, found by stepwise regression, in particular: median distribution of the height above ground (nZmedian) and fifth percentile of total pulse return intensity (i_tot5th). The former contained the most information and explained 58% and 62% of the variance in AGB and BA values; stepwise regression left us with two and four predictors, respectively, explaining 65% and 79% of the variance. For BA, the predictors were standard deviation, median and fifth percentile of total return pulse intensity (i_totstdDev, i_totmedian and i_tot5th) and nZmedian, whereas for AGB, only the last two were used. The plot-based metric showed that the median height of echo count (HOMTC) performs best, with very similar results as nZmedian, as expected. Cross-validation allowed the analysis of residuals and model robustness. We discuss our results considering our specific case scenario of a complex forest structure with a high degree of variability in terms of biomass.