Aims. Solar colors have been determined on the uvby-β photometric system to test absolute solar fluxes, to examine colors predicted by model atmospheres as a function of stellar parameters (T eff , log g, [Fe/H]), and to probe zero-points of T eff and metallicity scales. Methods. New uvby-β photometry is presented for 73 solar-twin candidates. Most stars of our sample have also been observed spectroscopically to obtain accurate stellar parameters. Using the stars that most closely resemble the Sun, and complementing our data with photometry available in the literature, the solar colors on the uvby-β system have been inferred. Our solar colors are compared with synthetic solar colors computed from absolute solar spectra and from the latest Kurucz (ATLAS9) and MARCS model atmospheres. The zero-points of different T eff and metallicity scales are verified and corrections are proposed. Results. Our solar colors are (b − y) = 0.4105 ± 0.0015, m 1, = 0.2122 ± 0.0018, c 1, = 0.3319 ± 0.0054, and β = 2.5915 ± 0.0024. The (b − y) and m 1, colors obtained from absolute spectrophotometry of the Sun agree within 3-σ with the solar colors derived here when the photometric zero-points are determined from either the STIS HST observations of Vega or an ATLAS9 Vega model, but the c 1, and β synthetic colors inferred from absolute solar spectra agree with our solar colors only when the zero-points based on the ATLAS9 model are adopted. The Kurucz solar model provides a better fit to our observations than the MARCS model. For photometric values computed from the Kurucz models, (b − y) and m 1, are in excellent agreement with our solar colors independently of the adopted zero-points, but for c 1, and β agreement is found only when adopting the ATLAS9 zero-points. The c 1, color computed from both the Kurucz and MARCS models is the most discrepant, probably revealing problems either with the models or observations in the u band. The T eff calibration of Alonso and collaborators has the poorest performance (∼140 K off), while the relation of Casagrande and collaborators is the most accurate (within 10 K). We confirm that the Ramírez & Meléndez uvby metallicity calibration, recommended by Árnadóttir and collaborators to obtain [Fe/H] in F, G, and K dwarfs, needs a small (∼10%) zero-point correction to place the stars and the Sun on the same metallicity scale. Finally, we confirm that the c 1 index in solar analogs has a strong metallicity sensitivity.