Background: Literature searches involve searching electronic databases, which requires a sensitive search strategy to capture as many relevant records as possible. Filters to identify test accuracy studies can lead to the omission of a considerable number of relevant studies. We have previously shown that searches designed to be very sensitive based on index test and target condition miss relevant studies indexed in MEDLINE. Objectives: To compare the performance of search strategies that do not incorporate a diagnostic filter (based on the index test and target condition) with the performance of searches that add a filter to these strategies. Methods: We included seven reviews that had each carried out extensive sensitive searches of multiple databases. We identified studies included in these reviews that were indexed on MEDLINE and these were used as our ‘reference’ set (523 studies). We ran searches of MEDLINE for each review based on the index test + target condition alone. We then combined these searches with each of 22 published diagnostic filters that we had translated to run on Ovid MEDLINE. We evaluated how many of the ‘reference’ studies for each review were identified by each of these 23 searches. We also looked at the number of records produced by each of the searches and their ability to reduce the number needed to read (NNR). Results: Searches designed to be very sensitive based on index test and target conditions miss an average of around 9% of studies indexed on MEDLINE (range 0% to 13% across reviews). Searches that included a filter missed on average between 14% and 58% of studies indexed in MEDLINE (ranged 3% to 88%). The NNR was reduced from 55 for index test + target condition searches to between 7 and 51 (median 27) for searches that incorporated a filter. None of the filters offer acceptable sensitivity for a reasonable decrease in the number needed to read. For moderate sensitivity (> 80%) the NNR was not substantially reduced (range 29 to 51). Conclusions: Currently available diagnostic filters should not be used to identify studies for inclusion in test accuracy reviews because they are unable to offer both acceptable sensitivity and a reduced NNR.