ABSTRACT Although many invertebrate animals orient by means of ultraviolet sky-light polarization patterns, existing measurements of these patterns are inadequate for full analysis of the biologically relevant information available from the sky. To fill this gap we have used a precision scanning polarimeter to measure simultaneously the intensity, degree, and direction of vibration (E-vector orientation) of polarized light at 5° intervals over the sky. The resulting sky maps were constructed for u.v. (350 nm) and visible wavelengths (500 and 650 nm) under a variety of atmospheric conditions. Our measurements confirmed that the patterns of radiance and degree of polarization of skylight are highly variable and hence unreliable as orientation cues; but patterns of E-vector orientation are relatively stable and predictable over most of the sky under all but very hazy or overcast conditions. The observed E-vector patterns correspond more closely to predictions based on first order (Rayleigh) scattering at 650 and 500 nm than at 350 nm. This is true both in terms of absolute accuracy and the proportion of the sky with relatively ‘correct’ information. Yet most insects respond to polarization patterns only at u.v. wavelengths. This apparent paradox can perhaps be resolved by assuming that there is no great selective advantage for any particular wavelength when large areas of blue sky are visible, but that under special and difficult conditions ultraviolet has advantages over longer wave-lengths. Measurements under partially cloud-covered sky, for instance, or under extensive vegetation, show that both spuriously polarized and un-polarized light resulting from reflexions present more troublesome interference at longer wavelengths than in the u.v. The accuracy of orientation achieved by dancing honey bees appears to be greater than can readily be accounted for by assuming that they use a strictly geometrical or analytical processing system for their orientation to polarized light.