Focal mechanisms of small earthquakes are critical for characterizing faults and regional stress. P-wave polarities and P/S amplitude ratios (e.g., HASH) are widely used to invert focal mechanisms for small earthquakes, but first-motion picking can be challenging, and P/S ratios are often largely inconsistent. In this work, I present FocMecDR, a relative focal-mechanism inversion method that utilizes double ratios of S/P amplitudes and relative polarities. Analogous to double-difference location and template-matching approaches, the method employs a template event with a known focal mechanism and, for nearby target events, minimizes the misfit between observed and theoretical P/S amplitude double ratios for event pairs while enforcing cross-correlation–based relative polarities. This empirical Green’s function–based strategy reduces sensitivity to absolute amplitude calibration and mitigates errors associated with first-motion polarity picking. Application to the 2019 Ridgecrest earthquake sequence demonstrates high accuracy for small events and indicates a relatively uniform stress field during the nucleation process. Using FocMecDR and 135 F-net focal mechanisms prior to the 2024 Mw 7.5 Noto earthquake as templates, reliable focal mechanisms are determined for more than 1,000 earthquakes with M > 2, delineating the dip angles of fine-scale fault structures and the spatiotemporal evolution of stress before the mainshock.