Detecting a target in clutter is particularly difficult because the observer must monitor many potential locations to find the target, and because the clutter itself might mask the target. To investigate whether contemporary models of search can account for visual search in clutter, we measured the detection of an oblique string of five aligned dots presented at an unknown location as a function of noise density. Observers judged which of two 200 ms intervals contained the signal string. At a given density, noise composed of oriented pairs of dots greatly degraded detection compared to random dot noise, especially if the paired noise shared the same orientation as the signal. Increasing the orientation difference between the paired noise and the signal improved detection, as did increasing signal length. We successfully modeled these results with an array of multi-scaled oriented detectors optimally tuned for the signal string. These results indicate that search for these simple patterns in noise is based on competing responses in oriented filters.