This article aims to examine and develop the corpus-based and statistical methods used in investigations of universal tendencies in translations, as well as to attempt to test and further revise the hypothesis of untypical lexical patterning in translations (Mauranen 2000). The focus is on synonymous words and their lexico-grammatical patterning in three subcorpora of Finnish Comparable Corpus of Fiction (FCCF). Synonymous items have been studied due to their interesting and problematic nature in translation studies already in both pre-corpus and corpus-based studies. The analyses are accomplished by applying a Three-Phase Comparative Analysis (TPCA), designed especially to analyse the source language influence. The TPCA and statistical procedures established that no consistent evidence for a universal untypical lexico-grammatical patterning could be found; rather they provided support for a source language dependent tendency. In a final instance, it is suggested that generalizations on translation universals must be done carefully since prior investigations seem to produce contradictory results, and the results of one study even show partly different tendencies.