Human–computer interface systems provide an alternative input modality to allow people with severe disabilities to access computer systems. One of the inexpensive and unobtrusive methods for this purpose is image-based eye blinks detection. Currently, available human–computer interface systems are often intrusive, limit in head rotation, require special hardware, and have special lighting or manual initialization. This paper presented a new robust method for real-time eye blinks detection. This method enables interaction using “blink patterns,” which are sequences of long and short blinks interpreted as semiotic messages. The precise location of the eye is determined automatically through multi-cues, accompanied by integration of eye variance feature and Gaussian Mixture Model classifier. The detected eye window is converted into a binary image. The eyelid’s distance is extracted by applying a variance projection derivative function. By following the eyelid’s distance in a finite-state machine, the blink patterns can be detected. The performance of the presented algorithm is evaluated using several frame streams. The experimental results show a robust eye blink pattern detection system in real environments.