International audienceThe development of an accurate analytical procedure for determination of dissolved water in complex alumino-silicate glasses via micro-Raman analysis requires the assessment of the spectra topology dependence on glass composition. We report here a detailed study of the respective influence of bulk composition, iron oxidation state and total water content on the absolute and relative intensities of the main Raman bands related to glass network vibrations (LF: not, vert, similar490 cm−1; HF: not, vert, similar960 cm−1) and total water stretching (H2OT: not, vert, similar3550 cm−1) in natural glasses. The evolution of spectra topology was examined in (i) 33 anhydrous glasses produced by the re-melting of natural rock samples, which span a very large range of polymerisation degree (NBO/T from 0.00 to 1.16), (ii) 2 sets of synthetic anhydrous basaltic glasses with variable iron oxidation state (Fe3+/FeT from 0.05 to 0.87), and (iii) 6 sets of natural hydrous glasses (CH2OT from 0.4 to 7.0 wt%) with NBO/T varying from 0.01 to 0.76. In the explored domain of water concentration, external calibration procedure based on the H2OT band height is matrix-independent but its accuracy relies on precise control of the focusing depth and beam energy on the sample. Matrix-dependence strongly affects the internal calibrations based on H2OT height scaled to that of LF or HF bands but its effect decreases from acid (low NBO/T, SM) to basic (high NBO/T, SM) glasses. Structural parameters such as NBO/T (non-bridging oxygen per tetrahedron) and SM (sum of structural modifiers) describe the matrix-dependence better than simple compositional parameters (e.g. SiO2, Na2O + K2O). Iron oxidation state has only a minor influence on band topology in basalts and is thus not expected to significantly affect the Raman determinations of water in mafic (e.g. low SiO2, iron-rich) glasses. Modelling the evolution of the relative band height with polymerisation degree allows us to propose a general equation to predict the dissolved water content in natural glasses: View the MathML source where CH2OT is the total water content (in wt%) dissolved in glass; TOTN represents the computed ILF/IHF variation as a function of the calculated NBO/T and SM parameters; IH2ON is the H2O band height scaled to ratio of the reference bands; k is the linearity spectrometer response on the H2OT band in function of water content. The water concentrations of the reference glasses are reproduced using this equation with a standard deviation of 0.06 wt%. The adopted parameterisation provides a useful tool towards the characterisation of composition dependence of micro-Raman procedures for silicate glasses. We show, based on the widest range of glass compositions so far investigated, that accurate evaluation of dissolved water content is achieved by micro-Raman spectroscopy