Abstarct: Electrical AC machines have been playing an important role in industry progress during the last few decades. Among the AC drives, permanent magnet synchronous machine (PMSM) drives have been increasingly applied in a wide variety of industrial applications. In many applications of PMSM, it is desirable to have an accurate control on Electromagnetic torque with a rapid response. Today one of the most prevalent torque controlled ac drives used for high performance applications are direct torque control (DTC) drives. The main features of DTC are: reduced torque oscillations, excellent torque dynamics, Absence of coordinated transformations (which are required in most of the vector-controlled drive implementations), etc. In the conventional scheme of DTC for PMSM there is no need for speed and position sensor. As stator flux lixnkages and electromagnetic torque are obtained by monitoring the stator currents and stator voltages, But for speed control purposes, a speed loop has to be added to the drive scheme and so speed information must be known. Usually the rotor position is measured by a shaft encoder, or Hall sensors. However, the presence of such sensors not only increases the cost and encumbrance of the overall drive system but also reduce its control robustness and reliability. Therefore, various position-sensorless control schemes have been developed for the estimation of rotor position and speed. This work presents a comparison between the most conventional methods for speed estimation in direct torque control of permanent magnet Synchronous motor such as Estimators baxsed on stator flux, and Model Refrance Adaptive system. Also a Speed Estimator baxsed on Neural Network has been introduced that has robust and adaptive estimation. It can be seen that each method has special advantages and disadvantages, and it is impossible to choose one of them as the best method. However, each method has priority over other methods depend on the condition.