Consider two small rectangular windows, each of which is centered at a different pixel in two different color images. Correlation is the process of computing the similarity between these two windows. If the camera rotation between the two images is not large then if these two pixels represent the same physical surface point then the similarity, and hence the correlation value, should be high. While correlation is a well known technique in computer vision, there are many possible variations. The main variability comes in the type of correlation algorithm, the colour space that is used, and the size of the window. This paper describes a set of experiments whose goal is to explore the impact of changing these three parameters. The methodology is first to manually select a set of matching features between two images, which are known as the ground truth. Then the different correlation variations are compared in terms of their ability to determine the correct matching feature out of all the possibilities.