Your webcam may know your face, but your keyboard knows your gender. Computer models can predict with 95.6% accuracy whether a man or woman is typing, according to a new study. To conduct the research, computer engineers installed keystroke-logging software onto the personal computers of 75 volunteers—36 men, 39 women—which monitored their daily computer use for 10 months. The researchers then used a program they created, called “ISqueezeU” to calculate the relative helpfulness of different typing features for determining gender—things like the time between two specific keystrokes, or the amount of time a key is pressed down during a single keystroke. A few features stood out as being more useful than others. For example, the average time between pressing the “N” key to pressing the “O” key was the most helpful, followed by the average time between pressing the “M” and “O” keys. The program isn’t capable of specifying whether a man or woman types those keys faster or more often—only that there is a difference. The researchers then tested the program’s findings using five machine learning models, which are computer programs that build models based on what they “learn” from existing data. All five models were able to predict gender accurately more than 78% of the time, with the most successful model being more than 95% accurate, the engineers report this week in Digital Investigation. The team proposes the use of keystroke dynamics as a cost-efficient and nonintrusive way to identify the gender of unknown computer users in criminal investigations, such as in cases of cyberstalking or identity theft. The researchers plan to expand their data collection with more volunteers, and see whether incorporating other variables such as handedness or education level can increase accuracy.