PurposeImage searching is a common activity for web users. Search engines offer image retrieval services based on textual queries. Previous studies have shown that web searching is more demanding when the search is not in English and does not use a Latin‐based language. The aim of this paper is to explore the behaviour of the major search engines in image retrieval using Greek text queries and to present and evaluate an image metaseacher that combines semantically similar queries to improve the relevance in image retrieval.Design/methodology/approachInitially the image retrieval capabilities (based on the number of items retrieved and their relevance) of search engines in Greek queries is studied with a number of semantically similar queries which differ in morphology. Then a system that produces semantically similar queries and merges their results is presented and the increase in relevance is measured. For the purpose of this paper, a number of queries suggested by a few students are run through the presented metasearcher and directly in the search engines. The participants of the evaluation study measured the precision in both cases.FindingsThe initial evaluation revealed that search engines retrieve different results in queries that differ in morphology or in grammar but still express exactly the same information need. Omission of diacritics affects the retrieval negatively as well. The study showed that the number of relevant images increases by combining the results of queries that differ in morphology.Originality/valueThe findings of this study could be applicable to other complex non‐Latin languages based, for example, on the Cyrillic alphabet. The presented metasearcher is a framework on how to expand the image retrieval capabilities of existing search engines. Its modular nature allows the straightforward integration of other techniques that are tailored to the characteristics of specific natural languages.