Scintillation crystals of 20 mm length or longer are needed for clinical time-of-flight positron emission tomography (TOF-PET) to ensure effective detection efficiency for gamma photons. However, the use of long crystals would deteriorate the key performance of TOF-PET detectors, time and spatial resolution, because of the variations in the travel times of the photons in crystals and the effects of parallax errors. In this work, we studied double-ended readout TOF-PET detectors based on coupling a long scintillation crystal to SiPMs at both ends for correcting the depth-dependent effects to improve the coincidence time resolution (CTR). In particular, we focused our attention to analyze timing performance using different correction methods, including trigger times of the individual photodetectors at both ends of the crystal, the simple average of the trigger times, and the weighted average based on the inverse variances of the depth-dependent corrected trigger times. For a 3 mm × 3 mm × 25 mm unpolished lutetium fine silicate (LFS) crystal with double-ended readout and practical head-on irradiation, a CTR of 246 ps FWHM can be achieved using depth-dependent timing-correction and weighted average time method compared to 280 ps FWHM using the conventional simple average time method and 393 ps FWHM using the conventional single-ended readout. The results show that the depth-dependent timing-correction and weighted average time method in double-ended readout can effectively correct for the trigger time variations in TOF-PET detector utilizing long unpolished crystals, resulting in an improvement in the CTR of as much as 37% compared to single-ended readout. © 2020 IOP Publishing Ltd and Sissa Medialab.