Abstract: Tenderness is one of the important assessment indices of beef quality. Traditional assessment methods, such as the sensory evaluation method and the Warner-Bratzler shear force method, have artificial error at different degrees. One steak from the mid-region of each longissimus dorsi (LD) was collected from each of 60 cattle as the testing sample. The age of cattle (400-550 kg) was from 30 to 36 months, and the cattle were fattened for more than 6 months on the same farm. After starving for 24 h, the live cattle were weighed, showered, stunned, killed, and letting blood for 56 min. After electrical stimulation, the 4 limbs and head of each animal were cut off, and the body of cattle was split into halves, cooled at 4℃ for 24 h, and then the carcasses were divided. The LDs were weighed, placed into plastic bags individually, vacuum-sealed, packed on ice, and transported to the laboratory. Each steak was cut into 10 cm×10 cm×10 cm samples, but the intermuscular fat and connective tissues were deleted. The samples were rinsed in water to remove surface contamination, then placed into plastic bags individually in a 75-80℃ water bath, and cooked for 15 min after the internal temperature of meat reached 70℃. Then the samples were cooled to room temperature (20℃). The 20 evaluators were healthy and dentally tidy adults with the age from 20 to 25 years old, without thirst or hunger. Each evaluator chewed the samples from each steak. After cooking, the samples within an LD were divided into 3 groups so as to run the experiments in triplicate. The samples freshly chewed for 0-20 times were measured using a Brookfield CT3 texture analyzer (Brookfield Engineering Laboratories, INC. Middleboro, Massachusetts, USA). With a two-cycle texture profile analysis (TPA) model and a TA44 probe (cylinder diameter = 4 mm), the size of testing surface of each sample was 10 mm × 10 mm × 10 mm. A Hold Time-pressure and keeping model was used throughout. The instrument settings were: pre-test speed of 2 mm/s, test speed of 5 mm/s, posttest speed of 5 mm/s, trigger force of 10 g, distance of probe movement on the sample of 7 mm, and hold time after downward movement of the probe of 2 s. For those samples, viscous force, stickiness, elastic force, elastic length, cohesiveness, resilience, gumminess, chewiness and other texture properties were measured using the texture analyzer. The correlations were analyzed between the parameters and beef tenderness level. The main texture properties decreased with the increase of beef tenderness grade, and the texture properties value also showed a downward trend with chewing more times. Combined with the sensory evaluation method, the BP (back propagation) network model, the RBF (radical basis function) network model and the self-organizing competition network model were built, and all the training errors were 1×10-6. Another steak from the mid-region of each LD collected from each of 20 cattle was selected as verification sample. Then the 3 network models were compared, and the self-organizing competition network model was the most accurate model with an accuracy rate of 90%, which showed that this method can accurately assess the level of beef tenderness.