Marketing research studies in India, particularly those based on consumer surveys, which extend data collection beyond a particular linguistic region, confront a serious problem of language comparability in the questionnaires if they are meant for administration to populations using different languages. The questionnaires are often developed in the English language and then are translated into appropriate language versions. All the data is put together without recognising the extent of measurement differences arising out of "language effect". The present paper provides an empirical test of this hypothesis. Consumer ratings of an advertisement was used as a situation in this study. A total of 130 words or phrases were used to represent 7 different dimensions on which an advertisement can be rated, namely, vigour, sensuousness, uniqueness, credibility, information content, irritativeness (or its reverse, attractiveness) and personal relevance. All these words were translated into a regional language and sample of 100 consumers was asked to rate the ad on these words first in one language and later in the other. It was hypothesised that the ratings in the two languages on any word should have equal means, high corelations and should produce comparable distributions. The hypotheses were statistically tested through analysis of variance Kolmogorov-Smirnov test for ranked categories and test of corelations. The results of this study are quite conclusive and indicate that nearly 50% of the words failed to produce a similar ratings at least on one of the three criteria when ratings in the regional language were compared with their English version. The study thus clearly establishes that the language effect seems to seriously affect the data collection process in market surveys and unless adequate care is taken in translations to make sure that resulting data would be comparable, a confounding in the result occurs without the researcher realising the magnitude of such confounding. This study recommends a serious pretesting of the translated versions of questionnaires and the use of play-back technique for reducing such confounding.