Abstract: Based on a monadic squarefree polynomial F（x,y） , which has roots x=φ i（y）,i=1,2,…,deg x（F）, where φ i（y）=C i,0 +C i,1 y+C i,2 y 2+… , a kind of four layer forward algebra neural networks with double inputs and single output and three layer forward algebra neural networks with single input and r outputs are designed , which can be applied to solve roots and approximate factorization. The neural networks learning algorithm is also designed. By using the learning algorithm, the divisors of F（x,y） and the number of divisors are decided, and the factorization of F（x,y） is realized.