The Long Range Reconnaissance Imager (LORRI) aboard New Horizons provides numerous optical images for precision photometry and astrometry, where accurate knowledge of the point-spread function (PSF) is essential. Existing Gaussian and effective-PSF (ePSF) models implicitly absorb pixel-integration effects, leading to pixel-phase-dependent centroid biases under near-Nyquist sampling. Following the super-resolution PSF reconstruction framework of Symons et al., we implement a pixel-integration–aware reconstruction of an oversampled stacked kernel with Richardson–Lucy deconvolution and iterative back-projection refinement to obtain a de-pixelated, band-limited PSF for LORRI (DePix-PSF). Tests on LORRI star-field images spanning magnitudes 6–16 show that DePix-PSF delivers nearly unbiased centroids and the smallest per-axis standard deviations of the O–C (observed-minus-calculated) residuals. On 1024 × 1024 data, it achieves centroiding precisions of 0.029 pixels (x) and 0.036 pixels (y), improving on the ePSF by 40% and 16%, respectively, and remains robust in single-exposure sparse-field cases where ePSF degrades. These results show that DePix-PSF measurably improves subpixel astrometry for LORRI. This method is also readily adaptable to other undersampled or near-Nyquist flight imagers.