A method for generating a single-cell three-dimensional image based on an optical flow analysis, comprising the steps: A. fabricating a light-induced dielectrophoresis chip (20), wherein the light-induced dielectrophoresis chip (20) is composed of a three-layer structure: the lower layer being ITO glass coated with a hydrogenated amorphous silicon coating layer, the upper layer being ITO glass without a coating layer, and a micro-fluidic channel being encapsulated between the upper and lower ITO glass layers for injecting a solution required for operations; B. injecting a cell and the solution into the micro-fluidic channel, and inputting a variable frequency alternating current signal to electrodes of the upper and lower ITO glass layers, and at the same time using incident light to irradiate the light-induced dielectrophoresis chip, so as to generate a non-uniform electric field in an irradiated area; C. changing the frequency and magnitude of the alternating current signal so as to control a cell motion direction, and at the same time collecting an image of the cell; and D. pre-processing the collected image, then performing feature extraction and velocity calculation, and finally reconstructing a 3D cell image. In the method, a cell two-dimensional image sequence is acquired by controlling the rotation of a single cell, and a maximum likelihood estimation method is used to realize a cell three-dimensional image generation technique based on a two-dimensional image sequence of a controllable single cell. The device used in the method is simple and low in cost, does not bring about a phototoxic function and has a relatively low requirement for a sample, thereby improving the operability.