BeiDou satellites transmit triple-frequency signals, which bring substantial benefits to carrier phase Ambiguity Resolution (AR). The traditional geometry-free model Three-Carrier Ambiguity Resolution (TCAR) method looks for a suitable combination of carrier phase and code-range observables by searching and comparing in the integer range, which limits the AR success probability. By analysing the error characteristics of the BeiDou triple-frequency observables, we introduce a new procedure to select the optimal combination of carrier phase and code observables to resolve the resolution of Extra-Wide-Lane (EWL) and Wide-Lane (WL) ambiguity. We also investigate a geometry-free and ionosphere-eliminated method for AR of the Medium-Lane (ML) and Narrow-Lane (NL) observables. In order to evaluate the performance of the improved TCAR method, real BeiDou triple-frequency observation data for different baseline cases were collected and processed epoch-by-epoch. The results show that the improved geometry-free TCAR method increases the single epoch AR success probability by up to 90% for short baseline and 80% for long baseline. The A perfect (100%) AR success probability can also be effortlessly achieved by averaging the float ambiguities over just tens of epochs.