An embodiment of the invention discloses a method and a device for diagnosing faults of meters in a flow measurement unit. The method comprises steps as follows: acquired reading jump values, readingtransfinite values and dead line times of a to-be-detected meter are subjected to normalization processing, acquisition values and dead line values are acquired, and the current reliability coefficient of the to-be-detected meter is acquired with a neural network algorithm on the basis of the acquisition values, the dead line values and the previous reliability coefficient of the to-be-detected meter, wherein during first input, the previous reliability coefficient of the to-be-detected meter is the preset maximum reliability coefficient representing normal operation of the to-be-detected meter; if the current reliability coefficient is smaller than a preset critical fault magnitude, alarming prompting information representing faults of the to-be-detected meter is generated, wherein the preset critical fault magnitude is the preset minimum reliability coefficient representing normal operation of the to-be-detected meter. The problem that fault conditions of the flow measurement unit inthe prior art are not timely mastered is solved, and the reliability of the meters in the flow measurement unit is improved.