An image processing apparatus, which quantizes M-level image data into N levels (M>N>2) using a multi-level error diffusion method or a multi-level average-error minimization method, includes a .gamma.-conversion unit that performs .gamma.-conversion on an integer part in a real number value after ideal gradation conversion that aims to obtain desired density or brightness, an area-gradation expressing unit that expresses a decimal part in the real number value by area gradation, an integerization unit that adds two values obtained through the .gamma.-conversion and the area gradation to obtain an integerized value corresponding to the ideal gradation conversion, a correction unit that adds a peripheral error to the integerized value, and a threshold setting unit that sets an N-1-type threshold value for a value corresponding to the ideal gradation conversion, wherein the conversion to N levels is performed by using the correction value and the N-1-type threshold value.