Cutting large thickness steel sheets faces a lot of limitations specially when using flame cutting such as burning surface and changing in the mechanical properties. Abrasive water jet machining (AWJM) can replace this method to avoid these limitations. This paper studies the most significant process parameters such as water pressure (P), traverse speed (Ts), and abrasive particles flow rate (Abr) as inputs on surface roughness (Ra), waviness (Wa), and kerf taper angle (Ө) as outputs when cutting steel 37 with 55mm thickness. The experiments were designed by Taguchi method. Taguchi analysis with signal to noise ratio (smaller is better) is used to determine the preferred levels of the input variables. A validation specimen is machined at the preferred levels which confirmed the correctness of Taguchi results. A prediction modelling using artificial neural network (ANN) is built between inputs and outputs to predict surface roughness, waviness and kerf angle for various inputs. The validation shows that the average total percentage error is 3.89%. The results show that the most significant parameter is the traverse speed on all outputs.