A neural network approach for on-line parameter estimation in unknown or poorly known processes with a time delay is proposed. The case of plants with unknown time delay and/or steady state gain has been considered. The main result of the paper is the analytical proof of the weight distribution as a sampling centred on the correct value of the time delay. Such a property, along with the estimation of the steady-state gain of the process from the sum of the weights, leads to an accurate identification of the unknown parameters of a process with time delay. A practical application of such a result is the design of an adaptive Smith controller. Simulation results are included in the paper to illustrate the proposed technique.