the fit improved by explaining an additional 2% of the total variability(Fig. 1). More importantly, this quadratic fit resulted in lower Aic values(53 vs. 73 for biovolume and 46 vs. 68 for biomass) and in an even distribution of the residuals, whereas in the case of the linear model the residuals described a dome-shaped distribution(data not shown). The presence of curvature in the size-scaling of metabolic rates was more evident when carbon-specific photosynthesis(PC) rates were plotted against cell using a semilog representation(Fig. 3). If a single, scale-free power law was capturing adequately the size-dependence of metabolic rate. the relationship between biomass- or volume- specific metabolic rate and cell size should display the ame slope throughout the entire cell size range. In contrast, we found that the slope of the P versus cell size relationship was positive in the small-to-medium cell size range and negative in the medium-to-large cell size range. This unimodal pattern. present in all growth phases. was particularly evident during the exponential growth phase