In the current issue of Apeiron [5(1-2), January-April 1998] Héctor A. Múnera looks at the non-null results of the entire corpus of Michelson-Morley class experiments: Michelson-Morley experiments revisited: Systematic errors, consistency among different experiments, and compatibility with absolute space. You will need an Acrobat reader, which can be obtained from Adobe. Here is the abstract:

Despite the null interpretation of their experiment by Michelson and Morley, it is quantitatively shown that the outcomes of the original experiment, and all subsequent repetitions, never were null. Additionally, due to an incorrect inter-session averaging, the non-null results are even larger than reported. Contrary to the received view, Illingworth's and other repetitions of the experiment were consistent with Miller's positive results. On the theoretical side, a new systematic error is uncovered: the angle between the projection of earth’s velocity on the plane of the interferometer and the reference arm of the apparatus has been practically ignored. This phase angle produces a noticeable change in the position of the peaks from one turn to the next of the interferometer. Hence, the data analysis cannot be based on the average of fringe shifts during a session, but rather on the calculation of individual speed for each turn. This procedure was applied to the only two sessions reported in detail in the literature: Miller's September 23, 1925 at 03:02 in Mount Wilson and Illingworth's July 9, 1927 at 11:00 in Pasadena. Surprisingly, it was found that in both cases the measured speeds exactly correspond to the projection of earth's orbital velocity only. As a result, the evidence against a preferred frame completely disappears.