As a test of the usefulness of matrix methods in extending phase information of proteins, the phases of triclinic lysozyme were calculated in the range from 3"3 to 2.5 A (2000 phases) using as a starting set the phases out to 3-3 A (1500 phases) and all the amplitudes. Observed ([Fobsl) and calculated (IFmodll) amplitudes were tested separately. The agreement of the predicted phases and the original ones was studied, and was observed to depend strongly on the normalized value of the corresponding structure factor, IEI. Different methods for assessing the quality of the predicted phases were considered, and the correl~ttion between electron density maps corresponding to predicted and original phases was selected as the most indicative of usefulness in structure determination. Predictions with the two different starting sets were studied using this correlation factor. A correlation of 0.5 between the predicted and original maps was obtained for the Fmodi, 0Crnodt case, and a correlation of 0.35 for the Fobs, ~,,od~ case. Both correlations are significant.