SUMMARYA method is presented for deconvolving collections of receiver functions at individual seismic stations. Conventional deconvolution trades-off signal enhancement (through stacking or simultaneous deconvolution) with preservation of directional variation. I replace stacking of traces with minimization of directional derivatives, thus communicating information between traces without stacking and recovering the minimum directional variation required by the data. Directional derivatives are defined by triangulation of the incident horizontal slownesses, computing derivatives at the centre of each triangle. Deconvolution is then posed as a simultaneous inversion for all traces, incorporating directional derivative minimization as cross-trace constraints. The constraints share frequency content between traces, while preserving directional variation. The resulting receiver function gathers have greater intertrace coherence, lower noise levels and higher usable frequencies than standard methods. A decade of teleseismic data at station GAC (Quebec, Canada) are deconvolved using conventional and panel deconvolution; the new technique produces a coherent receiver-function image and reveals complex crust and mantle responses. An extension of the method to recover harmonic coefficients of backazimuthal variation is also included.