A neuron is a linear combiner followed by a non-linear activation function. The activation function of the original ADALINE was a sharp quantizer that provided binary outputs for each input training pattern. Widely used in modern artificial neural networks is the more biological activation function called the sigmoid. A typical sigmoid function is the arc tangent. This provides an analog output that is bounded by +1 and -1. The sigmoid function is smooth and differentiable. At least squares training algorithm for a neuron with a sigmoidal activation function is derived. A trainable sigmoidal ADALINE is a basic building block of multi-layer neural networks.