In this research work, two recognition systems have been developed to recognize different breast tissues (normal or cancer infected). The multifractal dimensions have been used as textural discriminating feature. The difference between the two established systems is the first is based on using fuzzy logic to make recognition decision, while in the second system the feed forward neural network is used instead of fuzzy logic. The input to these systems is a breast tissue image. In this wok 24 type of breast images have been used, among this set one image represents the normal case (i.e., not cancer infected tissue image), while the other 23 images represent various types of breast tumors. At the beginning these images are analyzed using box-counting method to extract six fractal dimensions vectors. Then, these extracted feature vectors are used as input to the two recognition systems (fuzzy logic or ANN). In recognition system based on fuzzy logic, two types of membership functions (MSF) have been used (i.e., triangle and trapezoidal membership functions) to represent the extracted rules, which are applied on the FDfeatures. In recognition stage the sum of MSF for input samples is calculated and considered as recognition criteria, where class that shows the highest sum of MSF is considered as the class of the tested images. The experimental result showed that when value of number of bins parameter equal to 75 the recognition system show good performance in terms of recognition accuracy and processing time. Also, the result showed that the size of tested samples affects the recognition accuracy; when size of tested samples is large and near to the size used in training phase, then the rate of recognition becomes high. In artificial neural network (ANN) the 24 considered classes were divided into two classes: (i.e., either cancer infected or not infected). The training process is done by taking samples from all classes, and training the network using backpropagation NN. Then some testing is made using both the training and testing sets of samples. The test result showed that when using multi-fractal dimensions (with number of bins equal to 150) in training phase the recognition result is acceptable when both the training and test sets are used as test samples. The recognition results when using combined sets of fractal dimensions vectors are better than the results obtained by using one or two feature vectors only. The test result showed that the use of 4 hidden nodes was adopted because it makes the NN works with less computational time. Also, the learning rate equal to 0.5 led to best performance, because at this time the network required less training time with insignificant degradation security. Also, the test results showed that the size of tested samples doesn’t have significant effect on the recognition accuracy.