The aim of this paper is to investigate the morphology and syntax of numerals in Old Latvian according to a diachronic perspective. Several occurrences of numerals in the oldest texts (16th-18th centuries) were collected. As far as syntax is concerned, an alternation between the regular pattern of agreement and the use of a ‘common case’ is observed in numerals 1-9. Also ten and the other round numerals reveal a double behaviour, that is the partitive and non-partitive constructions. In spite of the ‘pervasive’ presence of a puzzling grapheme <e> in many endings, traces of the partitive model beside the non-partitive model are discovered in the oldest texts. Their distribution is accounted for by adopting a four-context pattern of analysis. Turning to morphology, I focused on the change from full forms to reduced and invariable forms. The data allow an update of the currently accepted chronology and a clarification of the gradual steps of this process consistently with the theory of the turning point.