I have coded a tanh-sinh quadrature routine (source at <[URL]http://www.codecodex.com/wiki/index.php?[/URL] title=Numerical_integration_with_tanh-sinh_quadrature>) which works, but the estimated error (from the last term) is not very accurate. I could try improving the error estimate by estimating the limit of successive approximations, but is there a more direct way of determining the error? Toby