Functional microarrays are powerful tools that allow the parallel detection of multiple strains at the species level and therefore to rapidly obtain information on microbial communities in the environment. However, the design of suitable probes is prone to uncertainties, as it is based so far on in silico predictions including weighted mismatch number and Gibbs free-energy values. This study describes the experimental selection of probes targeting subsequences of the nifH gene to study the community structure of diazotrophic populations present in Damma glacier (Swiss Central Alps) forefield soils. Using the Geniom One in situ synthesis technology (Febit, Germany), 2727 in silico designed candidate probes were tested. A total of 946 specific probes were selected and validated. This probe set covered a large diversity of the NifH phylotypes (35 out of the 45) found in the forefield. Hybridization predictors were tested statistically. Gibbs free-energy value for probe-target binding gave the best prediction for hybridization efficiency, while the weighted mismatch number was not significantly associated to probe specificity. In this study, we demonstrate that extensive experimental tests of probe-hybridization behaviour against sequences present in the studied environment remain a prerequisite for meaningful probe selection.