Abstract: A method of detecting the condition of a liquid layer level in a case where a plurality of liquids are in a layer condition, such method of being capable of knowing each layer level condition even in a case where a system as a whole is high at temperature, pressure or dirt as well as emulsification are large in extent. In order to detect the layer condition, the measurement values of temperatures and so on in a system as a whole or the mathematically processed values of the measurement values are provided as inputs, the layer level condition at this time is outputted, processed by a neurocomputer so as to compose an input-output model. As the layer level condition at this time is adapted to be estimated with the use of the input-output model, the layer level condition may be properly detected even when a system as a whole is high at temperature, pressure or the dirt as well as the emulsification are larger in extent.