PURPOSE: MR microscopy is in principle capable of producing images at cellular resolution (<10 µm), but various factors limit the quality achieved in practice. A recognized limit on the signal to noise ratio and spatial resolution is the dephasing of transverse magnetization caused by diffusion of spins in strong gradients. Such effects may be reduced by using phase encoding instead of frequency encoding read-out gradients. However, experimental demonstration of the quantitative benefits of phase encoding are lacking, and the exact conditions in which it is preferred are not clearly established. We quantify the conditions where phase encoding outperforms a readout gradient with emphasis on the detrimental effects of diffusion on SNR and resolution. METHODS: A 15.2 T Bruker MRI scanner, with 1 T/m gradients, and micro solenoid RF coils < 1 mm in diameter, were used to quantify diffusion effects on resolution and the signal to noise ratio of frequency and phase encoded acquisitions. Frequency and phase encoding's spatial resolution and SNR per square root time were calculated and measured for images at the diffusion limited resolution. The point spread function was calculated and measured for phase and frequency encoding using additional constant time phase gradients with voxels 3-15 µm in dimension. RESULTS: The effect of diffusion during the readout gradient on SNR was experimentally demonstrated. The achieved resolutions of frequency and phase encoded acquisitions were measured via the point-spread-function and shown to be lower than the nominal resolution. SNR per square root time and actual resolution were calculated for a wide range of maximum gradient amplitudes, diffusion coefficients, and relaxation properties. The results provide a practical guide on how to choose between phase encoding and a conventional readout. Images of excised rat spinal cord at 10 µm × 10 µm in-plane resolution demonstrate phase encoding's benefits in the form of higher measured resolution and higher SNR than the same image acquired with a conventional readout. CONCLUSION: We provide guidelines to determine the extent to which phase encoding outperforms frequency encoding in SNR and resolution given a wide range of voxel sizes, sample, and hardware properties.