[en] Fertility issues are a large part of economic losses for the dairy farmers. Early identification of pregnant and non-pregnant cows is a key element to improve reproductive performances and reduce costs for the farmer. The mid-infrared (MIR) spectrum obtained from milk recording routines is an inexpensive and quick method to obtain a fingerprint of the milk composition. This study was conducted in the context of the European project OptiMIR (INTERREG IVB North West Europe Program). The objective was to investigate the potential use of the entire milk spectrum to identify if a cow is pregnant or not. Investigation was based on 7,840 spectral records linked to confirmed pregnancy status coming from Luxembourg milk recording. The method was based on comparing a given spectrum to the expected spectrum if the cow would have been non-pregnant. The expected spectra were obtained from solutions of a mixed model (fixed effects: parity, herd, milking moment and days in milk; random effects: animal across lactations) applied to MIR spectra from a subset of non-pregnant cow. Therefore the solutions obtained in the model were used on the whole dataset to obtain predicted MIR spectral values for all test-days and prediction errors (residuals) representing the factors not present in the model (reproductive status, unaccounted factors, and error). A predictive quadratic discriminant function was then constructed on the residual spectra to predict the pregnancy status. Leave one out cross-validation showed promising results with an error rate equal to 1.8% and 6.8% for non-pregnant cow and for pregnant cow respectively. Results have shown that MIR milk spectra might be used as a pregnancy diagnosis tool. Therefore, this kind of diagnosis could be made routinely and at a low cost for farmers.