Performing tasks require us to abide and apply the relevant rules to the task (ie., do not wash white clothes with coloured clothes). As we perform tasks, these rules are maintained through cognitive control processes. Functional magnetic resonance imaging (fMRI) studies have shown that rule representations are decodable in a set of regions called the multiple demands (MD) network. Another set of regions called the default mode network (DMN) decorrelate with task-related networks. However, recent studies show that rule representations are decodable in tasknegative networks such as DMN as well. The present study investigated rule decodability using multivariate fMRI analysis methods. Subjects performed an event-related rule switch fMRI experiment in which there were two rules: categorizing numbers as even/odd (parity rule) or greater/smaller (value rule) than 10. In our first analysis, these two rules were modelled based on the switch or repeat status of the specific trial. We found that on switch trials, rules were decodable across the whole brain. In contrast, on repeat trials rules could not be decoded from any brain region. Curiously, repeat trials rule classification was significantly below chance, suggesting that training the classifier on one set of patterns made them worse than chance in classifying on the remaining set. This would happen if different instances of the same rule on repeat trials may have been accompanied by very different patterns of activity. In our second analysis we investigated if rule-related patterns changed with the number of consecutive repetitions or the number of consecutive switches, e.g., if the activity-pattern related to the parity rule on the first instance of a repeat trial was different from the pattern when this rule repeated again. We found that this was indeed the case. While the same rule could not be classified much across two consecutive switch trials, the same rule could be classified nearly everywhere across two consecutive repeat trials.