This study explores collocations in mechanical engineering texts by applying a multilingual contrastive analysis approach. The point of departure in the contrastive analysis is the morpho-syntactic properties and the lexical realization of the semantic level, viewing a collocation as a unit of meaning and as an equivalent or translation unit. British contextualism (Firth, Sinclair) and lexicological approaches (Hausmann, Cowie) were adopted as two major theoretical approaches, but, considering various distributional, semantic, conceptual, and pragmatic aspects of collocations, further theoretical approaches were needed that draw on theories of lexical semantics and corpus and computational linguistics methods. In this study, collocation is understood as a typical combination of mainly two words characterized by habitualness, high frequency of co-occurrence of its constituents, and their specific semantic hierarchical relationship (Hausmann, 1984, 1995). The aim of the study is to describe the specific properties of nominal collocations identified in the English sub-corpus (adjective + noun, noun + noun), to compare them with their equivalents in German and Croatian, and to highlight the most significant features of this sample as well as the observed universals. Applying a corpus-based two-tier methodological model, collocations were extracted from the English sub-corpus, serving as a starting sub-corpus, by use of the computer software AntConc, in particular the Cluster/N-Grams function and the concordance tool (Anthony, Windows 3.5.0.). Once a frequency-sorted list was created from each monolingual sub-corpus, a list of collocation candidates was created by applying a frequency threshold of 5, which provided data for the fine-grained quantitative and qualitative analysis. To describe the relationship of collocation constituents and to clarify the creation of meaning by their cooccurrence in a syntagmatic sequence as well as by the interdependence of their meanings (‘meaning by collocation’, Firth, 1957), typical association measures, popular in computational linguistics, were used (the mutual information, the t-score, the chi-square test, the loglikelihood ratios, and P). The comparison of the morpho-syntactic properties and semantic relations of identified equivalent collocations in German and Croatian revealed five different structural patterns, and a very high level of full lexical correspondence. The adjective + noun collocation pattern is found to be a multilingual equivalent, whereas nominal compounds proved to be an exclusive feature of German, and noun + prepositional phrase pattern of Croatian. The common matching pattern of German and Croatian is noun + genitive noun. According to these findings, the structural collocation patterns in the German and Croatian languages depend on the language type and the possibilities allowed by word formation processes. Since the subject of the research is LSP collocations, their terminological status, role in the hierarchical structure of taxonomies and the relationship of equivalence in a contrastive terminological analysis were also concerns of interest here. Phenomena observed by contrasting them within the multilevel framework were presented systematically and constituted the basis for the derivation of universals by inductive reasoning. The findings of the empirically collected data analysis confirmed the proposed four hypotheses. In the semantic relation established by the realization of collocation constituents’ meaning potential, the collocate generates the specialization of the base meaning by actualizing one of its meanings, and at the same time contributes to the semantic transparency of the collocation. Given the intra-collocational cohesion, and (non)substitutability of collocation constituents as one of the major features of collocations, the substitution of components is possible only with synonymous lexemes retaining the semantic value of the collocation. Variants of terminological collocations were confirmed in all three sub-corpora exhibiting remarkable typological diversity and uneven representation. Furthermore, an important distinctive feature established cross-linguistically is the prevalence of nominal compounds in German as equivalents of English collocations demonstrating high equivalency at the lexical, semantic, conceptual, and functional levels.