Long baseline interferometer observations have shown that the relative positions of strong components in the water source W49(H2O) change by angles of about 0.05 arcsec in periods of a few months (Knowles et al.). Bodily motion of the individual components cannot explain the apparent changes, since this would imply velocities about 10,000 km/sec, which are much too high; a model in which a number of differently situated masers switch independently on and off is statistically improbable. Refraction effects producing 'ghost images', of the type suggested by Gold, cannot account for the position jitter. Neither, it appears, can inherent instability in the solution of the radiative transfer equations, postulated by Deguchi. It is hard to explain the variations of position by stochastic motions of irregularities within the masing clouds, so it seems most likely that changes in the location of the pump energy input to the maser are responsible for the observed behavior.