In this doctoral thesis, computational models for the assessment of physico-chemical properties of mineral insulating oils based on data collected using molecular spectroscopy techniques were developed. The goal of this thesis was to develop computational models which will predict physico-chemical properties of oils from their molecular spectra with a high level of confidence, and through which information is provided on oil ageing and any chemical impurities present. Along with model development, the relationships between physico-chemical properties of oils were investigated to contribute to a better understanding of information provided by the tests. Analyses of significant spectral regions in molecular spectra were performed, signals which vary with changes in physico-chemical properties were determined, and the responsible chemical species were identified. Many samples (1135 samples) of insulating oils of different ageing levels were prepared. Part of the samples consisted of completely new oils of different chemical composition, another part was prepared from new oils through a method of accelerated ageing at elevated temperatures and under oxidizing conditions, and the last part consisted of real samples from electrical equipment in operation. Physico-chemical properties were tested on all the samples (colour, acidity, interfacial tension at the oil-water interface, density, dielectric dissipation factor, and specific resistivity) and molecular spectra were measured for specific samples (infrared, ultraviolet-visible, and Raman spectra). Based on the measured physico-chemical properties, correlation curves were made between different properties. Relationships were investigated between properties that are indicators of the chemical composition of oils, between properties which are indicators of oil ageing, as well as interdependences between properties which describe oil composition and oil ageing. Since molecular spectra consist of a very large number of data points that have different chemical meanings and describe chemical bond vibrations and molecular electronic transitions, the significances of spectral regions for the assessment of each individual physico-chemical property of insulating oils were determined using the variable importance in projection method and the selectivity ratio method. With the known significances of spectral regions, regression models were developed for the assessment of physico-chemical properties of oils using partial least squares regression modelling as well as artificial neural network modelling. Infrared spectra were used to develop models which, with a very high degree of confidence, which is within the repeatability criteria of standardised methods, predict the colour, acidity, interfacial tension at the water-oil interface and density of mineral insulating oils. The acidity, colour, and interfacial tension of the oil correlated highly with the carbonyl spectral band, and it was concluded that carbonyl compounds, more specifically carboxylic acids, aldehydes, and ketones are responsible for the changes in properties of insulating liquids during ageing. Changes in colour of insulating liquids are visible to the naked eye, and the compounds responsible are quinones, which are cyclic unsaturated ketones. Density, which first and foremost depends on the chemical composition of the oil, shows the greatest correlation with vibration bands which describe the hydrocarbon composition of the oil, more precisely the characteristic vibrations where alkanes and alkenes can be detected, as well as the intense C–H stretches at higher wavenumbers. The interfacial tension of oil is the only property which shows a high degree of dependence on both chemical composition and oil ageing. Assessment models for dielectric dissipation factor and specific resistivity of oils from infrared spectra were not successful, and the reason was impossibility of quantification of chemical compounds which contribute to those properties (micrometre size particles of metals and polymers, carbon black, and water) with infrared spectra. Based on ultraviolet-visible spectra, a computational model which estimates the colour of insulating liquids with a very high level of confidence was developed. The colour of oil can be predicted due to the coloured nature of quinones formed during oil ageing, which absorb in both infrared and ultraviolet-visible spectra. It was not possible to estimate other physico-chemical properties from these spectra due to the impossibility of monitoring the content of carbonyl compounds and the hydrocarbon composition of oils with ultraviolet-visible spectra. Raman spectra were unusable for computational model development due to very low signal strength, high levels of noise during measurement, and the appearance of luminescence which masks signals in aged samples. It is possible that the use of differently configured Raman spectrometers and other laser sources would provide better data. The results presented in this thesis imply the possibility of using spectroscopic assessment methods in supplementing the existing laboratory measurement methods, as well as in the development of simple measurement devices which could monitor the physico-chemical properties of insulating oils in the field or during equipment operation.