Maloney (1986) reported that 3 basis functions, derived from principal components analysis of the reflectances of Munsell papers, can account for >99% of the variance (from black) in a set of natural reflectances. This raised the possibility that a trichromatic visual system can generate accurate estimates of the reflectances in natural visual scenes. To address this question, a set of 377 reflectance spectra from natural objects (leaves, flowers, and fruits) was collected. (Note that Krinov's reflectances, used in Maloney's and others' studies, represent the space-averaged reflectances from large natural formations, not the reflectances of individual colored objects). The responses of human cones to these natural surfaces was calculated using the cone sensitivities from Stockman et al. (1993). Because these cone sensitivities are not necessarily related to the principal components of the reflectances, a Cone-Based Linear Model (CBLM) was developed for this analysis (R. O. Brown, 1993 OSA Annual Meeting). The CBLM provides a least-squares best fit to the natural reflectances, using the 3 cone responses as coefficients. The CBLM was compared to 2 linear regression models, which are based on the 3 principal components from either the Munsell reflectances (LR-M) or the natural reflectances (LR-N). All these analyses assumed a constant, known illuminant (CIE Source C), and covered the range 400-650 nm.