This paper discusses in practical terms the problems which can affect research in the area of product optimisation and flavour development, as viewed from the standpoint of an agency researcher specialising in food and drink research. The conventional approach to product development tends to rely on consumer tests, such as product placement or taste tests, to identify perceived strengths and weaknesses of the product via rating scales and spontaneous comments. This information is used by technical personnel to guide product development, and one or two new formulations are subsequently tested and, if 'better', they are progressed. At best this approach will identify a possible product improvement, but at its worst it can be unproductive as technical personnel struggle to achieve viable improvements. The weakness underlying this approach is that it assumes a direct relationship between a product formulation and the consumer response to it: there is no such relationship. I will go on to explain why this is so. I argue, therefore, that the use of consumer opinions, as expressed via rating scales and spontaneous comments, to guide product reformulation, is unsatisfactory. We should instead ask the consumers to do only what they are really competent to do: to say how much they like something, and perhaps to say which product they prefer. This determines that we must identify the factors to assess in the research and to test these systematically and in a controlled way. I will go on to give examples of how this can be done. It should be noted that I will not attempt to discuss specific aspects of product testing, such as the difference between blind and branded tests, nor the merits of various types of semantic rating scales. The argument I present is concerned with product development and as such does not touch on other types of research such as concept and placement tests - within these limits, however, I feel the ideas I present are valid across a range of product types and situations, and do not depend on very specific aspects of detail research design.