Deoxynivalenol (DON) and its modified forms (3-, and 15-acetyl-DON, DON-3-glucoside) are commonly analysed by chromatographic methods. Indeed, coupled with proper extraction and clean-up, LC-MS represents the best approach for multi-mycotoxin measurements. On the other hand, immunochemistry-based methods are possibly able to detect a family of structurally related compounds, although the determination of single contributions is not possible so far. However, ELISA methods often lead to an apparent overestimation of the mycotoxins content because modified forms and matrix components can potentially cross-react with the antibodies (designed for the parent toxin). Several data about the possible cross-reactivity of commercial DON-detecting ELISA kit are reported in the literature so far. Data are commonly obtained in buffer solutions or in matrix-matched solutions, but comparison of a set of naturally incurred samples has never been reported. In the present work the accuracy of a commercial DON-detecting ELISA kit was evaluated on naturally incurred soft wheat (n = 15) and maize (n = 15), taking into account the matrix effect. Recovery was calculated considering the DON concentration found by LC-MS/MS and the total DON concentration, expressed as the sum of DON and its modified forms found by LC-MS/MS. The obtained data clearly show that, when 3-modified forms of DON occur in the sample, the ELISA kit does actually detect them, thus returning an apparent overestimation if only DON content is considered. When the ELISA recovery is calculated on the total DON content, the accuracy of the analysis increases and the variability decreases. According to our data, the ELISA kit seems to be a promising group detection tool for the accurate evaluation of DON and its modified forms, expressed as sum of DON, DON-3Glc and 3Ac-DON, for soft wheat and maize samples.