This study shows that cause and types of errors in complex problem-solving tasks can be explained within a framework of the prevalence effect commonly studied only in simple visual search tasks. The explanation proposes that subjects make a series of probabilistic decisions aimed at balancing both speed and accuracy. Such decision is a complex process that relies not only on task instructions but also on cognitive biases established by the history of previous trials and progress of the current trial. We provide evidence based on both empirical data and cognitive modeling.