We identify essential features of accelerometer and gyroscope data that reflect the movements of tendons (passing through the wrist) when performing a finger or a hand gesture. With these features, we build a classifier that can uniquely identify 37 (13 finger, 14 hand and 10 arm) gestures with an accuracy of 98\%. We further extend our gesture recognition to identify the characters written by the user with her index finger on a surface, and show that such finger-writing can also be accurately recognized with nearly 95% accuracy. Our presented results will enable many novel applications like remote control and finger-writing-based input to devices using smartwatch.