In geophysics the signal resulting from tidal forcing is the only one precisely known a priori. Its omnipresence in tilt measurements can be exploited to characterize the spatio-temporal properties of rocks in situ but only if experimental uncertainties are less than 2%. Signal distortions due to unknown material and geometrical discontinuities around sensors limit the observational accuracy. Tilt tides were measured with a new biaxial tiltmeter installed in a 12 m borehole at Leduc, Alberta, where the geology is well defined both locally and regionally. At the M2 frequency, the observed parameters (error of 2–5%) are coherent with results previously obtained with near-surface tiltmeters installed 100 m away but 5 times noisier. The load tide due to oceans differs according to which configuration is used in its computation, and the new measurements favor Schwiderski's M2 model over that of Parke and Hendershott. A topographic perturbation due to a small valley at 500 m from the borehole has been identified by comparing the tilt residuals for M2 and O1 with the strain tide. The results indicate that even a smooth topography can perturb tidal observations to the point of hampering the interpretation of tilt anomalies in terms of regional tectonics.