In this paper an adaptive threshold logic unit is analyzed. The unit consists of a set of self-adjusting weights, a summing device, and a comparator. Its dynamic and steady-state behavior is made clear by investigating the solutions of a system of nonlinear differential equations which describes changes in the weights. In the steady state the unit separates its inputs into a family of principal components and picks out one principal component for its output. When the power of one principal component is considerably larger than that of any other, it picks out the largest-power principal component without regard to initial values of the weights. When all the principal components have nearly the same values of power, any principal component can be given as the output and the initial conditions determine which should be given. These theoretical results are verified by some simulation experiments on an analog computer. Applications of the unit to many-valued logic and signal filtering are also discussed.