Optimal sensor placement is essentially a decision problem under uncertainty. The maximum expected utility theory and a Bayesian linear model are used in this paper for robust sensor placement aimedat operational modal identification.To avoid nonlinear relations between modal parameters and measured responses, we choose to optimize the sensor locations relative to identifying modal responses.Since the modal responses contain all the information necessary to identify the modal parameters, the optimal sensor locations for modal response estimation provide at least asuboptimalsolution for identification of modal parameters. First, aprobabilistic model for sensor placement considering model uncertainty, load uncertainty and measurement error is proposed.The maximum expected utility theory is then applied with this model by considering utility functions based on three principles: Shannon information, quadratic loss, and K-L divergence.In addition, the prior covariance of modal responses under band-limited white-noise excitation is derived and the nearest Kronecker product approximation is employed to accelerate evaluation of the utility function. As demonstration and validation examples, sensor placements in a 16-degrees-of-freedom sheartype building and in Guangzhou TV Towerunder ground motion and wind load areconsidered. Placements of individual displacement meter, velocimeter, accelerometer and placement of mixed sensors are illustrated.