The technological advances in computing and the development of corpus linguistics have revolutionized the methodology of diachronic linguistics with undeniable benefits in terms of speed of data collection and volume and diversity of these data. Unlike the philological method, based on reading and rereading a text, computerized corpora accessed with automatic search tools provide in seconds a large amount of samples taken from hundreds of works that the investigator may not have read thoroughly. The lack of deep familiarity with both the content and the language system of the texts cannot affect the grammatical interpretation of many fragments, but in relevant cases can be misleading, as we illustrate with some examples. The cautious analysis of unexpected records and its extensive contextualization before venturing an interpretation can help us overcome this obstacle.