When submitting queries to information retrieval (IR) systems, users often\nhave the option of specifying which, if any, of the query terms are heavily\ndependent on each other and should be treated as a fixed phrase, for instance\nby placing them between quotes. In addition to such cases where users specify\nterm dependence, automatic ways also exist for IR systems to detect dependent\nterms in queries. Most IR systems use both user and algorithmic approaches. It\nis not however clear whether and to what extent user-defined term dependence\nagrees with algorithmic estimates of term dependence, nor which of the two may\nfetch higher performance gains. Simply put, is it better to trust users or the\nsystem to detect term dependence in queries? To answer this question, we\nexperiment with 101 crowdsourced search engine users and 334 queries (52 train\nand 282 test TREC queries) and we record 10 assessments per query. We find that\n(i) user assessments of term dependence differ significantly from algorithmic\nassessments of term dependence (their overlap is approximately 30%); (ii) there\nis little agreement among users about term dependence in queries, and this\ndisagreement increases as queries become longer; (iii) the potential retrieval\ngain that can be fetched by treating term dependence (both user- and\nsystem-defined) over a bag of words baseline is reserved to a small subset\n(approxi-mately 8%) of the queries, and is much higher for low-depth than deep\npreci-sion measures. Points (ii) and (iii) constitute novel insights into term\ndependence.\n