This article introduces the KneE-PAD, which is a dataset consisting of knee rehabilitation exercises performed by 31 patients suffering from knee pathologies. In particular, a total of 267 patients were monitored over a 6-month period where they were asked to perform in physiotherapy centers without any supervision 3 common lower limb rehabilitation exercises (squats, leg extension and walking). At each participant a set of 8 sEMG and IMU sensors were placed at important lower limb muscle groups. After curating and grouping the wrongly executed exercises, 2 common wrong variations for each exercise were identified in 31 participants, while a total of 2,086 files are available with each one having an approximate duration of 4.2s. The goal of KneE-PAD is to be used for training machine learning algorithms for automatic postural assessment using only wearable sensors (sEMG and IMU), which could become a vital part of a virtual coach to supervise the patients and provide useful feedback to them while executing their prescribed rehabilitation exercises remotely.