### abstract ###
The problem of classifying sonar signals from rocks and mines first studied by Gorman and Sejnowski has become a benchmark against which many learning algorithms have been tested
We discovered that both the training set and the test set of this benchmark are linearly separable, although with different hyperplanes
Moreover, the complete set of learning and test patterns together, is also linearly separable
We give the weights that separate these sets, which may be used to compare results found by other algorithms
### introduction ###
It has become a current practice to test the performance of learning algorithms on \textsl{realistic} benchmark problems
The underlying difficulty of such tests is that in general these problems are not well caracterized, making it thus impossible to decide whether a better solution that the one already found exists
The Sonar signals classification benchmark, introduced by Gorman et al CITATION  is widely used to test machine learning algorithms
In this problem the classifier has to discriminate if a given sonar return was produced by a metal cylinder or by a cylindrically shaped rock in the same environment
The benchmark contains 208 preprocessed sonar spectra, defined by  SYMBOL  real values, with their corresponding class
Among these,  SYMBOL  patterns are usually used to determine the classifier parameters through a procedure called learning
Then, the classifier is used to class the  SYMBOL  remaining patterns and the fraction of misclassified patterns is used to estimate the generalization error produced by the learning algorithm
We applied Monoplane, a neural incremental learning algorithm, to this benchmark
In this algorithm, the hidden units are included one after the other until the number of training errors vanishes
Each hidden unit is a simple binary perceptron, trained with the learning algorithm Minimerror  CITATION
