A method is presented for off-line calculation of optimal measurement subsets to be used during implementation of optimal feedback control for linear stochastic systems. It is seen that feedback control imposes a penalty on noisy measurements which must be included with measurement cost in the optimization. The combinatorial problem involved with subset selection is avoided by creating a non-linear ‘ optimal control ’ problem which can be solved using available algorithms. Results for a scalar example show that, with costly measurements, it is most economical to use a number of short measurement periods.