Affirmation and negation is two main semantic and grammatical categories in any language. In order to explain the processing mechanism of negative sentence, some researchers have proposed schema-plus-tag model and two-step simulation hypotheses. However, these hypotheses cannot explain the processing mechanism of isolated negative sentences with uncertain states, such as the skirt is not blue. So anchor-based activation and satisfaction-constrained model was proposed to solve this problem. In the present study, we used eye-tracking to give further support to this new model. In the three experiments, a visual-world paradigm was adopted to explore the processing mechanism of negative sentences with uncertain states under three different contexts. In Experiment 1, participants were only presented with the alternative choice of the depicted event (e.g., red skirt), short for ‘A’; In Experiment 2 only presented ‘NOT X’ (e.g., blue skirt attaching negative marker); In Experiment 3 presented both ‘A’ and ‘NOT X’ simultaneously. In these three eye-tracking experiments, participants were first presented with the sentences by voice, and then at 1200 ms, four pictures were presented simultaneously at the beginning of the state word (e.g., blue). And participants’ task was to choose which picture matched the sentence. The materials of Experiment 4 were adopted from Experiment 3, however, presented by offline. So, participants have plenty of time to choose the matching pictures. The results demonstrated that, at the early stage of processing, participants had higher fixation probabilities to pictures depicting the negated state of affairs, such as ‘blue skirt’ in all the three experiments. Then at the later stage, participants showed higher fixation probabilities to the pictures depicting ‘A’ in Experiment 1; however, to ‘NOT X’ in Experiment 2 and 3. In addition, participants showed lower fixation probabilities to the pictures depicting the negated states than the random level. In Experiment 4, 30 out of 37 participants chose the pictures depicting ‘A’ rather than ‘NOT X’ being the matching pictures. The results from all four experiments showed that, when processing the negative sentences with uncertain states, participants would first simulate an anchor (representation of the negated state of affairs), then searched for the actual state of affairs based on the given context. If participants had enough time to get the actual state of affairs based on linguistic features and the given context, then the searching stopped and accepted the actual state as the final state of affairs. However, if participants cannot get the actual state of affairs based on linguistic features and the context or had not enough time to finish the whole process, then the searching also stopped and accepted the representation of the negated state of affairs attaching negative marker being the processing result. At the later stage, the representation of negated state of affairs would be suppressed for useless. The results support anchor-based activation and satisfaction constrained model as well as the suppression hypothesis.