Actually I am not 'loading' the internal referance by using it for outer ciruit. for external ckt i have an other 2.5V referance.

rather typical amount of noise, actually quite good.

yes , but actually i forgot to tell that, i am taking 100 samples in a second, summing the all 100 times adc counts. when 100 samples completed, i derive the average by deviding the sum figure by 100. and then this average value is dispalyed.

hence each second the dispaly is updated, i expect not more than +-2 counts variation when the input is constant.