When observers are asked to report a feature of a single target displayed in rapid serial visual presentation, they frequently make errors. Most frequently, a feature from the to-be-reported dimension pertaining to a stimulus presented near the target is reported. These migrations are so-called illusory conjunctions in the time domain. From parallel models, it is proposed that during the presentation of the series, the response features of the stimuli are extracted. If a high rate of presentation does not enable proper binding processes, the system could base its response on sophisticated guessing on the basis of the relative levels of activation of the available response features. However, the multiple extractions assumption has not received direct empirical support. We report two experiments in which the observers had to report their first and second response candidates. This double response paradigm makes it possible to test the assumption that more than one response feature is available for making a response. Furthermore, the application of the constant ratio rule (following Botella, Barrio pedro, & Suero's, 2001, model) to the first responses allows predictions for the ratios between choices of the items for the second responses. The correlations between the observed and the predicted response proportions were .887 and .956 in the two experiments. This high predictive capacity indicates, first, that the observers have more than one response available, among which to choose, and second, that the choice among responses is determined largely in the same way for both first and second responses. Nevertheless, the small errors in prediction are further reduced if a proportion of pure guesses is assumed in the second responses. These are probably due to memory losses, misidentifications of the features, and other factors impairing performance in second responses in comparison with first responses.