Extreme learning machine (ELM) for single-hidden layer feedforward neural networks (SLFNs) was known for its extremely fast learning speed while maintaining acceptable generalization. Unfortunately, the failure of ELM on big data occurs frequently. The course is, the main computation of ELM focus on the calculation of generalized inverse of hidden layer output matrix, which depends on singular value decomposition (¡SVD) and has very low efficiency especially on high order matrix. In view of this high calculation complexity directly courses the failure of ELM on big data, normal equation extreme learning machine is proposed, which use the normal equation to reduce the size of the matrix equation and overcome the failure. The experiments on benchmarks show that the new proposed model has better performance than the ELM, so as to have more potential for large scale data learning.