This paper is devoted to the problem of nonlinear state estimation under multirate sampling in presence of disturbance inputs. Considering a general description of a nonlinear sampled-data system, we establish a prescriptive framework for multirate observer design via an approximate discrete-time model of the plant. This framework is shown to be input-to-state stable in a semiglobal practical sense with respect to the estimation error for the unknown exact discrete-time model. A numerical example of an aerospace vehicle with input and output channels of various sampling rates demonstrates how the multirate observer can drastically improve performance compared with the single-rate observer.