A partial flow blockage in an assembly of a liquid metal reactor could result in a cooling deficiency of the core. To develop a partial blockage detection system, we have studied the changes of the temperature fluctuation characteristics in the upper plenum according to changes of the t10w blockage conditions in an assembly. We analyzed the temperature fluctuation in the upper plenum with the Large Eddy Simulation (LES) turbulence model in the CFX code and evaluated its statistical parameters. Based on the results of the statistical analyses, we developed a neural network model for detecting a partial flow blockage in an assembly. The neural network model can retrieve the size and the location of a flow blockage in an assembly from a change of the root mean square, the standard deviation, and the skewness in the temperature fluctuation data. The neural network model was found to be a possible alternative by which to identify a flow blockage in an assembly of a liquid metal reactor through learning and validating various flow blockage conditions.