Three methods for second language data collection are compared: free composition, picture description and translation. The comparison is based on percentage of errors in a grammatical category and in a source category. Most results obtained from the free compositions and picture descriptions tended to be similar. Greater variation was found for some error categories between these two tasks and the translation task. Analysis of the errors suggests that differences in results could be reduced through slight adjustments in the method of data analysis, and a variation of the translation task. Results obtained from the three methods should then be very similar. (Author)