Masking of the non-test ear is often needed in speech audiometry but the methods for such masking have not been well defined. 'White' or 'speech' noise is commonly provided by audiometers for this purpose, and the many problems and uncertainties in their calibration and effectiveness are discussed; data on these aspects are presented with respect to some current audiometers and Fry's and AB(S) PB-word lists. Formulae are given for estimating the possible need for masking, calculating the required dial level of masking noise for a given dial level of speech signal, and assessing the extent to which cross-masking may be affecting the results. The formulae presented are applicable for earphone listening only.