This work exploits the possibilities offered by the recently developed multivariate method named Science-Based Calibration (SBC), for the extraction of 'analyte-specific' chromatograms in on-line gradient reversed phase LC-infrared spectrometry (IR) in the presence of a high spectral and chromatographic overlapping between the analyte of interest, co-eluting sample matrix constituents and the mobile phase components. The SBC method uses an experimentally measured single response spectrum of the analyte of interest and representative noise to calculate an optimum regression vector (b(opt(1))). Then, the b(opt(1)) vector is used to predict the concentration of the analyte of interest in the spectra of the LC-IR sample chromatograms. To evaluate the advantages and pitfalls of the proposed approach, two different situations were analysed on real LC-IR data sets obtained from the injection of a series of standard solutions of four nitrophenols (p-nitrophenol, 3-methyl-4-nitrophenol, 2,4-dinitrophenol and 4-nitrophenol) in a reversed phase system under gradient conditions. In the first situation, the extraction of the 'analyte-specific' chromatogram was carried out without previous knowledge of the spectral features of other interferents present in the sample matrix. In a second situation evaluated, data obtained from the LC injection of a sample blank is available. Results show the potential applicability of this technique in a variety of situations and evidenced that the proposed chemometric approach improves the selectivity and sensitivity of the LC-IR hyphenation.