The invention discloses a network flow thread level dynamic load balancing method and system based on two-stage forwarding. The method comprises the following steps: setting a controller, and setting a state monitoring Agent on each server; 2) triggering a collection module to send a collection request to CPU core operation state information of an Agent collection server in each state collection period by the controller; 3) calculating the weight of each CPU core by a calculation module according to the CPU core operation state information; 4) releasing the weight of each core to a first stage forwarding device by an execution module, and if the current stream is a new stream, allocating a destination CPU to the current stream according to the weight of the core, then filling a server sequence number to which the destination CPU belongs and a CPU core sequence number in an MAC address of outputting the flow, and updating the flow table; if the current stream is an existing stream, forwarding the existing stream to the destination CPU of the previous forwarding; and forwarding the flow to a destination port by a second stage forwarding device according to the server sequence number in the MAC address.