Information of relevance to the organism often is represented by the concentrations of particular molecules inside a cell. As outside observers we can now measure these concentrations precisely, but the cell's own mechanisms must be noisier. We show that, in certain limits, there is a universal tradeoff between the information capacity of these concentration measurements and the amount of relevant information that is captured, a version of the information bottleneck problem. This universal tradeoff is confirmed, quantitatively, in an analysis of the positional information encoded by the gap genes in the developing fly embryo.