A recent evaluation of the effect of a Nepalese language teacher training program on student language learning is described and discussed, focusing on the difficulties faced by external evaluators working in difficult circumstances and not on study results. The program provided inservice training for secondary school teachers of English as a Second Language, using locally trained Nepali staff and an expatriate training officer. The evaluation study involved comparison of language performance of students taught by program-trained teachers with that of students taught by untrained teachers. Baseline measurements of student performance were taken at grade 8 in 12 experimental and 12 control schools, and posttest results were ultimately available from 11 experimental and 11 control schools. In addition, teachers were observed in class, were interviewed, wrote reports of their lessons, and submitted samples of pupils' work. Selection of participating schools was heavily influenced by serious problems of communication and information gathering in Nepal. Methodological problems were found in the following areas: the design of language tests used; quantity and reliability of data gathered; potential for bias in the observation structure; use of external evaluators; and sampling. Some recommendations are made for improving the process. (MSE)