<p>Zircon studies in mafic and ultramafic rocks are particularly valuable because they can reveal deep-mantle petrogenetic and geodynamic processes. These studies, however, are hindered by the scarcity and uneven distribution of zircon in the mantle and mantle-derived rocks. Consequently, finding zircon in these rocks is inherently difficult and strongly influenced by the zircon abundance, grain size, volume, and the number of samples searched. Here, we describe a method for in-situ zircon identification and extraction for further analytical studies. To this end, we cut four representative 2 × 3 cm slabs, one from the granite and three from the gabbro, and carefully polished one side to enable automated large-area elemental mapping with SEM, EPMA, and µ-XRF using WDS and or EDS. Of all these, µ-XRF offers the best balance among acquisition time, surface coverage, spatial resolution and result quality. To overcome spectral interferences affecting the Zr signal, the method uses multidimensional analysis based on suitable ratios of the X-ray lines produced by the sample's major and, eventually, minor elements. Zircon-bearing pixels are then discriminated utilising a combination of global, local, and probabilistic statistical classifiers, such that zircon identification is accepted only where consistent statistical behaviour is reproduced across independent methods. Overall, the protocol enables reproducible detection and extraction of zircon down to ~ 50 µm in grain size while minimising both false positives and false negatives. The same methodology can be used to locate other Zr-bearing (e.g., baddeleyite, zirconolite, srilankite) and Zr-lacking (e.g., apatite, xenotime monazite, rutile, chromite) accessory minerals that may be present in the rock sample.</p>