This paper examines the methodological challenges of identifying literary book reviews in newspapers,contrasting manual and automated approaches. By discussing the manual, and ’traditional,’ approach ofa previous literature study alongside computational methods for classifying book reviews, we explore how human and automated approaches provide complementary perspectives on this task with a focuson the National Library of Sweden’s historical newspaper collection. Along with different findings, the paper highlights key issues related to the arbitrariness of ‘what constitutes a book review,’ digitisation and annotation issues and differences between frequency-based and BERT methods. We conclude by suggesting that nuanced text mining of specific types of newspaper articles benefits from considering both contextual and computational perspectives, which can together enhance our understanding of the complexities involved.