The quality of knitted fabric in circular knitting machines is highly sensitive to any undesired changes in the mechanism and components involved. For instance, a broken needle causes defects on the surface of knitted fabric. Consequently in order to increase the quality and reduce production cost, rapid detection and diagnosis of defected needles on industrial circular weft knitting machines is a crucial need. In these machines when the yarn is pulled down by the needles to knit a loop the created yarn tension, causes fluctuations in the feeding yarn flow. The aim of present research is to identify broken needle defects and their numbers, during yarn feeding in a circular knitting machine, employing neural network analysis on yarn fluctuation signals. The experiments procedures were designed so that three needle defected conditions were implemented on an industrial circular knitting machine. The yarn fluctuation signals were captured and saved, then using wavelet the contaminated signal noise was removed. Statistical and wavelet analysis are implemented to produce the required features. Finally the capability of neuro network for classification of four groups of data including healthy, one, two and four broken needles were examined. The results show that 99.43 % accurate distinction of broken needles is achieved in 50 iterations.