Nonuniformities of sound level occur in all reverberant rooms including those used in sonic testing. A theory is presented which describes these variations statistically in terms of analysis bandwidth and room reverberation time. The theory applies above a certain cutoff frequency and in regions remote from room boundaries. These nonuniformities produce uncertainty in the room averaged sound spectrum levels leading to potentially serious risk of undertest or overtest. A methodology is proposed whereby the risk of any degree of undertest or overtest can be predicted and controlled in advance. The methodology employs hypothesis testing applied to the known statistics of the sound field.