Forty participants compared pairs of adjacent positive and negative numbers which were associated with temperature values by using the comparative instructions “Colder?” and “Warmer?”. In this experiment, the stimulus pairs were presented horizontally and responses were made using left and right manual key presses. The presence of a significant four-way interaction between absolute digit magnitude, digit sign, the form of the comparative instruction, and the side of response provided a very challenging set of constraints for RT modelling. Nonetheless, a good fit to these data could be obtained within a random walk model framework by using adjustments to the drift rates based on both correspondences between polarity codes (Proctor & Cho, 2006) and the nature of the semantic congruency of the pairs with the comparative instructions (Leth-Steensen & Marley, 2000).