The potential of decoding handwriting trajectories from brain signals for use in brain-to-text communication has yet to be fully explored. Here, we developed a novel brain-computer interface (BCI) paradigm that tried to fit the trajectories of imaginary handwriting movements from intracortical motor neural activities and translate them into texts using machine learning approach. The trajectories for handwriting of digits and multi-stroke characters were decoded using a diverse array of neural signals, achieving an average correlation coefficient of 0.75. We developed a speed profile identifier based handwriting recognition algorithm, which accomplished a recognition rate of around 80% within an extensive database of 1000 characters. Additionally, our research uncovered a notable distinction in the neuronal direction tuning between writing strokes and cohesions (air connections between strokes), leveraging which a dual-model approach could exploit to enhance performance by up to 11.7%. Collectively, these findings demonstrated a new approach for BCIs that could possibly implement a universal brain-to-text communication system for any written languages.TeaserHandwriting trajectory was successfully decoded from brain signal for direct brain-to-text translation of any written languages.