The origin of pseudotachylytes has been controversial since Wenk cast doubt on the melt origin of the matrix of pseudotachylytes, in 1978. The matrix of this rock is so fine grained that the crush origin of pseudotachylytes, revived by Wenk, cannot easily be denied. This paper presents a new line of argument based on the size analysis of clasts contained in pseudotachylytes in felsic granulite from the Musgrave Range, central Australia. These clasts are definitely crush products produced during the pseudotachylyte generation. Their sizes, as measured on photomicrographs of thin sections, obey the size distribution: N = N′r-D, where N is the cumulative number of clasts with sizes greater than r, D is the fractal dimension, and N′ is a constant that depends on the number of measurements. D was found to be 1.5 ± 0.05 for the size ranges of 10-2000μm. If the matrix of the pseudotachylytes consists mostly of ultrafine crush products, they must have formed simultaneously with those coarse crush products. The proportion of fine products relative to the coarse clasts can be estimated, assuming that a similar size distribution also holds for the fine products. The estimated area occupied by fine products in a thin section is of the order of only several percent, whereas the measured area of the matix is about 60%. Thus the major part of the matrix of the pseudotachylytes cannot be regarded as crush products. It is also shown that the number of clasts smaller than about 5 μm becomes very small, perhaps as a result of nearly complete dissolution of fine clasts in a melt. However, if the ultrafine-grained matrix of the pseudotachylytes had formed by crushing during seismogenic fault motion, the grain-size refinement during the crushing should have occurred jumping the size range of at least 1-5 μm. This is quite unreasonable and disproves the crush origin for the matrix of pseudotachylytes.