Problems of transcribing Greek children's speech - as well as the speech addressed to them - through the computational programs of the Child Data Exchange System (CHILDES) are discussed. Regardless of how much they involve technical or theoretical issues, all such problems are argued to have repercussions on research analysis and conclusions. Transcription ones are particularly acute in Greek, as the program relies on latin characters rather than the International Phonetic Alphabet. In fact, decision as to how to transcribe phonemes like θ or dynamic word stress, are the result not only of technical biases but also of research priorities, more specifically in our case the devising of a lexicon for child speech. In this sense, even the original transcription line of the program can not be taken as a neutral base, which can simply be enriched by more specialized analyses and, as others have claimed, transcription is theory. Coding of data was for various reasons restricted to an elementary syntactic categorization of the lexicon. Yet, two serious problems were faced, regarding what can be taken as a word and how it can be coded grammatically. It is, for instance, frequently difficult to decide what to do with interjections, amalgams of all shorts, clitics and neologisms with no conventional meaning. Syntactic categorization of lexemes also encounters the problem o early words not being yet or fully grammaticalized.