This study describes the requirements and methods for textizing statistical tables using OCR (optical character recognition) at scale. A major challenge of textizing statistical tables using OCR is analyzing the table layout with high accuracy. I develop a Python tookit, ocrstats, which supports the task by providing batch processing, automation of routine processes, use of external OCR, and table layout analysis with practical accuracy. I also explain the practical tips learnt from the process of textizing the Japan Imperial Statistical Yearbook using ocrstats.