Limits to sensory detection are always at some (physical or neural) level set by signal/noise. In vision, photon statistics sets one absolute limit to the information that can be extracted from an image, but the information may be partitioned differently depending on the needs of an animal in a particular situation. Primarily, a balance has to be struck between sensitivity and resolution, secondarily between different kinds of resolution: spatial, temporal and chromatic, with polarization as an additional dimension in some species. Realizable combinations of these depend on the anatomical and physiological implementations in each species. As an example, I shall summarize our studies on temporal integration (which favours sensitivity) vs. temporal resolution in humans and amphibians. In rod vision, the balance is set at the very input: the time scale of rod phototransduction limits all temporal aspects of scotopic vision. Cone vision, however, appears to be slower than cone phototransduction would allow. Modelling suggests that "unnecessarily" fast cone signals in humans may be useful for intra-retinal signal processing that serves to enhance spatial resolution under small eye movements.