The goal of conjunctive visual search is to attentionally selecta location at which the visual array matches a set of cued fea-ture values. Here we present a neural dynamic architecturein which all neural processes operate in parallel in continu-ous time, but in which discrete sequences of processing stepsemerge from dynamic instabilities. When biased competitionselects an object location at which not all conjunctive featurevalues match the cue, the neural representation of a conditionof dissatisfaction is activated and induces an attentional shift.Successful match activates the neural representation of a con-dition of satisfaction that ends the search. The search takesplace in the current visual array but takes into account an au-tonomously acquired feature-space scene memory.