The purpose of this project was to design an algorithm for detection of tonic seizures based on surface electromyography signals from the deltoids. A successful algorithm has a future prospect of being implemented in a wearable device as part of an alarm system. This has already been done for generalized tonic-clonic seizures, and the hypothesis was that some of the same characteristics could be found for tonic seizures. The signals were pre-processed by a high-pass filter to remove low frequency noise such as movement artifacts. Several different features were investigated, including kurtosis, median frequency, zero crossing rate and approximate entropy. These features were used as input in the random forest classifier to decide if a data segment was from a seizure or not. The goal was to develop a generic algorithm for all tonic seizures, but better results were achieved when certain parameters were adapted specifically for each patient. With patient specific parameters the algorithm obtained a sensitivity of 100% for four of six patients with false detection rates between 0.08 and 7.90 per hour.