A general theory of signal-to-noise ratio (SNR) in simultaneous acquisition of spatial harmonics (SMASH) imaging is presented, and the predictions of the theory are verified in imaging experiments and in numerical simulations. In a SMASH image, multiple lines of k-space are generated simultaneously through combinations of magnetic resonance signals in a radiofrequency coil array. Here, effects of noise correlations between array elements as well as new correlations introduced by the SMASH reconstruction procedure are assessed. SNR and SNR efficiency in SMASH images are compared with results using traditional array combination strategies. Under optimized conditions, SMASH achieves the same average SNR efficiency as ideal pixel-by-pixel array combinations, while allowing imaging to proceed at otherwise unattainable speeds. The k-space nature of SMASH reconstructions can lead to oscillatory spatial variations in noise standard deviation, which can produce local enhancements of SNR in particular regions.