A sensitive and fast HPLC method using ultraviolet diode-array detector (DAD)/electrospray ionization tandem mass spectrometry (Q-TOF-MS/MS) was developed for the determination of tebipenem pivoxyl and in the presence of degradation products formed during thermolysis. The chromatographic separations were performed on stationary phases produced in core–shell technology with particle diameter of 5.0µm. The mobile phases consisted of formic acid (0.1%) and acetonitrile at different ratios. The flow rate was 0.8mL/min while the wavelength was set at 331nm. The stability characteristics of tebipenem pivoxyl were studied by performing stress tests in the solid state in dry air (RH=0%) and at an increased relative air humidity (RH=90%). The validation parameters such as selectivity, accuracy, precision and sensitivity were found to be satisfying. The satisfied selectivity and precision of determination were obtained for the separation of tebipenem pivoxyl from its degradation products using a stationary phase with 5.0µm particles. The evaluation of the chemical structure of the 9 degradation products of tebipenem pivoxyl was conducted following separation based on the stationary phase with a 5.0µm particle size by applying a Q-TOF-MS/MS detector. The main degradation products of tebipenem pivoxyl were identified: a product resulting from the condensation of the substituents of 1-(4,5-dihydro-1,3-thiazol-2-yl)-3-azetidinyl]sulfanyl and acid and ester forms of tebipenem with an open β-lactam ring in dry air at an increased temperature (RH=0%, T=393K) as well as acid and ester forms of tebipenem with an open β-lactam ring at an increased relative air humidity and an elevated temperature (RH=90%, T=333K). Retention times of tebipenem pivoxyl and its degradation products were used as training data set for predictive model of quantitative structure–retention relationship. An artificial neural network with adaptation protocol and extensive feature selection process was created. Input parameters for model were calculated from molecular geometries optimized with application of Density Functional Theory. The model was prepared and optimized especially for small data sets such as degradation products of specific compound. Validation of the model with statistical test against requirements for QSAR showed its ability for prediction of retention times within given data set. Mean error of 24.75% (0.8min) was achieved with utilization of topological, geometrical and electronic descriptors.