Tiltmeters consisting of a servoed bubble trapped under an optical flat are described, together with techniques used for calibrating, orienting, and installing such instruments. The long-term stability of one such (quartz) instrument is such that a 2-μrad annual tilt of the Poorman Hill can be measured; however, we cannot tell whether an apparently steady tilt of 3 μrad/yr is real or due to instrumental drift. A beryllium copper instrument drifted at a more rapid but relatively constant rate of 22 μrad/yr. The two instruments agree at the 107-rad level over a 2-month interval, after a linear term has been removed from both records. Earth tide tilts observed with the quartz instrument before and after construction of a concrete block enclosure, and with the beryllium copper instrument at two different sites in close proximity to the quartz instrument, all differ in amplitude and phase. The difference between a pair of observations at a single tidal frequency approximates a periodic unidirectional tilt with maximum excursion near the time of greatest earth strain. For a given pair of observations the direction is the same at all frequencies, but this direction differs for different observation pairs. These differences are ascribed to local tilts generated by tidal strains acting on the geometrical irregularities of the observing site. Tilts due to ocean loads and coupling of strain with local geology and topography have been computed, but it is not possible to calculate the very local site effects owing to the complex and irregular nature of the site. Strain-induced tilts are believed to be the major source of error in tilt measurement: the larger-scale couplings are probably calculable in most cases, but the very local site effects must be avoided by the use of long-base instruments in either tunnels or boreholes.