Pre-election surveys to predict election outcomes have produced inconsistent results because different researchers or research organisations apply different methodologies and research designs even when dealing with the same subject. The object of this article is to interrogate the nature of survey research as a scientific tool in general and to trace the development of its use in political and government settings, especially in South Africa, to predict election results. The study strives to establish whether or not some pre-election surveys in South Africa rely on inadequate methodology, or else succumb to political pressure to produce predictions that favour a certain party? Attention is directed to issues of reliability and the validity of survey results that can predict the outcomes of elections with a high degree of accuracy. We examine the extent to which different survey organisations and researchers heed the prerequisites and demands of scientific methods in research and, more especially, the methodologies used in surveys. Examining results produced by various research organisations we explore whether political surveys today are adequate tools to predict scientifically outcomes in an election or whether they are simply mechanisms used to arrive at desired goals at the expense of scientific methods. Common flaws in methodologies used to make predictions are identified and discussed.