Advancements in Brain-Computer Interface (BCI) technology offer to the users the ability to control external devices without physical movement. Electroencephalography (EEG) based BCI systems have been gaining attention for their superior temporal resolution, ease of use, and portability. While the potential of EEG for decoding complex body movements, such as finger movements, is considered crucial for daily activities, research focused to improve the high spatial resolution of EEG remains limited. Conventional EEG systems, following to the standard 10–20 electrode distribution and typical mounting structures, often suffer from low spatial sensor resolution. Our study pioneers the use of newly designed flexible electrode grids directly applied to the scalp to collect ultra-high-density EEG (uHD EEG) recordings. In the present work, a total of 256 channels densely distributed over the contralateral sensorimotor cortex was used. Notably, the adopted dense electrode distribution and small-sized electrodes allow for an inter-electrode distance of 8.6 mm, in contrast to the average 60–65 mm spacing in conventional EEG systems. A novel, deep learning-based approach for uHD EEG processing and classification was developed in the present work. Motor imagery trials of individual fingers’ movements during uHD EEG recording were carried out. The performance of our novel approach was assessed providing very good accuracy. Our findings may shed a new light on the potential of uHD EEG for enhanced precision in decoding complex motor actions, promising advancements in BCI technology for real-world applications.