Summary Episodic tremor and slip events in Northern Cascadia produce low-frequency, emergent seismic signals, often referred to as tectonic tremor. Methods designed to locate tremor face two challenges that increase the likelihood of producing low quality catalogues: first, signal arrival and duration are often poorly defined; second, high rates of tremor activity during episodes can cause multiple signals to be indiscernible from each other. In this paper, we present a new method of locating tectonic tremor using differential traveltimes from a waveform-envelope cross-correlation in a grid-based Bayesian inversion. To address the aforementioned challenges, we use a recently-developed three-dimensional shear wave model to compute traveltimes, and include processes to remove data outliers, estimate data error statistics, and quantify uncertainties within the Bayesian framework. Although this method is designed for tremor, to test the approach we first consider a set of 58 local earthquakes between magnitudes −0.07 and 2.6 in the Southern Vancouver Island region and obtain well-constrained relocations. Residuals between official catalogue values and our relocations are quantized with respect to the 1 km grid resolution of the inversion, and average 2.7 km in epicentre and 5.2 km in depth. Analysis shows that depths of relocations are sensitive to horizontal variations and simplifications in velocity models. We then present our catalogue of tremor events during the 2004 episodic tremor and slip event beneath Southern Vancouver Island, Canada. Median uncertainties of tremor events quantified by 95% credibility interval widths in a 1 km grid are 5 km and 9.5 km in horizontal and depth directions, respectively (1.2 km and 2.3 km using traditional standard deviation-based uncertainties). Comparison of our catalogue with previously published work demonstrates that our new method yields a good detection rate, a greater degree of epicentral clustering, and better depth resolution of tremor events. Catalogues produced using this new method may help to provide insight into the spatial extent of tremor, especially in depth, by yielding enhanced constraints on source locations on a regional scale.