Changes in tilt have been measured by annual or semi-annual spirit level surveys of small-aperture (40 to 400 m) bench mark arrays located at borehole tiltmeter sites along the San Andreas Fault in California. Because there are six or more bench marks in most arrays, the tilt is overdetermined, and realistic estimates of uncertainties can be made. The large arrays (aperture 300 m) afford a precision (two standard deviations) of about 3 μrad in measuring tilt, and the measured tilt remains constant to within ±5 μrad over periods of up to 5 years. The small arrays (aperture <100 m) afford a precision of about 6 μrad in measuring tilt, but the measured tilt exhibits a high variability significant even at that level of precision. The lower precision of the small arrays is primarily a consequence of minor bench mark instabilities and local short-wavelength elevation disturbances (root mean square variability perhaps 0.25 mm), but the high variability of tilt measured by the small arrays appears to be a product of local intermediate-wavelength (50 to 300 m) elevation disturbances (perhaps due to the mechanisms suggested by Harrison, and Harrison and Herbst) rather than true tectonic tilting. In general, borehole tiltmeter recordings show large changes (>10 μrad) in tilt within periods on the order of a year, but the best tiltmeters exhibit a stability comparable to that obtained from surveys of the large-aperture bench mark arrays.