Determination and in situ detection of bearing radial clearance are studied using vibration spectra detected by acceleration sensors applied to double-row self-aligning ball bearings. We applied three different methods for the determination of bearing clearance and present the results: a) using neural networks, b) calculating the spectral kurtosis of the corresponding spectra and c) performing recurrence plots and recurrence quantification analysis for various bearing clearances. The statistical results and corresponding quantificators show reliable in service detection and monitoring of the clearance.