In recent decades, motor imagery (MI) based brain computer interface (BCI) is frequently used to control the external devices, but it has limited applications because of its lower classification accuracy (CA). The classification accuracy depends on the feature extraction techniques which is a challenging problem in the field of MI based BCI systems. In this paper, an efficient feature extraction and classification technique is proposed for the detection of multi-class (left hand, right hand, tongue and foot) MI movements with improved classification accuracy. The proposed phase space reconstruction (PSR) technique extracted efficient features during MI activities and the support vector machine (SVM) classifier was used to classify multi-class MI movements. The proposed technique and the classifier are tested on BCI competition-III (2005) dataset-IIIa which contains four-class of MI movements of the subjects. The results showed that the proposed technique improved the classification accuracy of MI signal and has better performance (%CA = 80.86% and Cohen's kappa coefficient (K)= 0.72) compared to several state-of-the-art techniques. © 2019 IEEE.