Non-invasive brain-computer interface (BCI) systems typically work with electroencephalography (EEG) signals, which, due to their small amplitude, can be naturally contaminated with artefacts. Artefacts can modify the EEG signals and obscure or be misinterpreted as neurological events, diminishing the reliability of the EEG and posing a risk of compromising BCI control. Thus, attenuating EEG artefacts is essential. For BCI systems, the method of artefact attenuation should be automatic, online, and ideally performed with few EEG recording channels. A method for eyeblink artefact reduction based on adaptive filtering which satisfies all these conditions was proposed in previous work. The present study aims at investigating the reliability and effectiveness of the technique in different conditions through extensive tests and evaluating its performance in simulations mimicking the real environment. The results were compared among the simulated scenarios. After incorporating causality and non-linearity into the artefact, the filter maintains good performance even after the introduction of these features. The lowest overall RMS errors were achieved using a 20th-order filter and adaptation factor set in 10–7 for both scenarios. The proposed method resulted in a significant artefact reduction, preserving ERP (event-related potential) morphology, and allowing easy component identification even when the blink overlaps the desired ERP. All this is performed in an online and automatic fashion, with few EEG recording channels and without the need for a reference channel, hence being a great choice as a pre-processing step to attenuate eyeblink artefacts in BCI systems.