This paper describes a shape recognition technique using boundary chain codes extracted by a method as described by Pavlidis and used an 8 neighbourhood. A chain code is a representation of a two-dimensional contour using a one array. Feed forward neural networks were recognise these chain codes. In addition, backpropagation network is trained u training algorithms and the resulting optimal parameters are recorded. Depending upon the complexity of the object to be recognised, this technique can used to form the basis for object recognition or as the best method. The research is also aimed to compare the performance of chain code representation as against centroidal profile extraction. The third objective is to determine the effectiveness of Feed forward artificial neural networks recognising and classifying different medica items in the image form. The networks were trained on a large number of medical waste items. The wide variety of shapes and textures revealed that just a representation of an object's boundary is not sufficient to recognise every object in the set, some form of texture recognition will also be required in recognising medical wastes. The results have shown that chain code has lesser performance as compared to centroidal profile representation.