Two benchmark arrays located on the islands of Santo and Efate in the New Hebrides Island Arc have been releveled 2 to 4 times per year during the past five years (1975–1980). Each array is thereby used as a multicomponent tiltmeter. The arrays are located within 35–50 km of the interplate thrust zone of this convergent plate boundary. Each array has an aperture of about 1 km and in addition contains small subarrays with apertures of about 70 m. This configuration leads to a resolving power (for tilt) of about one microradian. The most important result of this study is that, at the Devils Point array on Efate Island, there has been a consistent tilting down of about five microradians in a direction subparallel to the strike of the arc. This tilting is clearly resolved by both the large array and by the small aperture sub-arrays. The observations are accounted for by a combination of planar tilting and random error consistent with the observed misclosures. Although tilting is clearly established at the Devils Point array, it is not possible at this stage to establish a unique cause for the observed tilt. Hypotheses consistent with the tectonic setting include interseismic strain accumulation associated with the earthquake cycle of loading and interplate boundary rupture, deformations near a transition between a creeping and a locked segment of the interplate boundary, and near surface deformations associated with magmatic activity and/or block movements also reflected in the pattern of faulting of the uplifted Holocene coral terraces. At the second array at Ratard (Santo Island) no tilt trend clearly emerges from the noise level. However, two marginally significant tile excursions precede in each case large earthquakes located several hundred km north of the array. At the Ratard array the ground noise (small wavelength elevation disturbances and benchmark instability) contributes to residuals almost as much as the measurement error, thereby degrading the threshold level for detection of tilt signals. The difference between the two arrays is not clearly established but seems to reflect an inherently larger level of short wavelength ground noise associated with geological properties of the Ratard site.