AbstractSerial femtosecond crystallography (SFX) at X-ray free electron laser (XFEL) facilities enables structure determination of radiation sensitive systems and transient intermediates in protein dynamics at room temperature. However, it also poses significant challenges for data collection and data processing. Crystal heterogeneities and beam instabilities result in shot-by-shot variation of diffraction images. The high amount and rate of raw data make visual evaluation by humans impossible and data processing a tedious task. Thus, an automated data processing pipeline is essential to make strategic decisions during data collection and make the most of limited XFEL beamtime. A data processing pipeline for SACLA, an XFEL facility in Japan, was developed. Cheetah was modified to retrieve images directly from the data acquisition (DAQ) system and parallelized over many cores on many nodes. A graphical user interface was also developed to control jobs and visualize the results. The pipeline consists of two stages. The online stage provides realtime feedback on the hit rate and detector saturation. This is essential to assess crystal density, beam alignment and attenuator thickness. The offline stage outputs hit images in a compressed, multi-event HDF5 file together with metadata. The outputs are processed by CrystFEL. The pipeline is routinely used for more than 90 % of SFX experiments at SACLA, including experimental phasing and time-resolved studies. In the second part of the talk, experimental phasing at SACLA will be discussed. In addition to already published results on Hg-SIRAS, Cu-SAD and S-SAD phasing, more examples will be showcased, including (1) phasing from less than 3,000 indexed patterns, (2) a novel compound to derivatize membrane proteins and (3) S-SAD phasing of a membrane protein in lipidic cubic phase. In collaboration with Helen Ginn at Oxford, we discovered that postrefinement can further reduce the number of necessary images by nearly half.