The assert_allclose text is not precise enough to be helpful to fix a test failure that cannot be replicated on every machine, and we cannot just quickly grab --pdb-failures. By how much do I have to lower the precision to make it pass on this continuous integration machine? assert_allclose(he, hefd, rtol=5e-10) File "C:\Python27\envs\py3\lib\site-packages\numpy\testing\utils.py", line 1297, in assert_allclose verbose=verbose, header=header) File "C:\Python27\envs\py3\lib\site-packages\numpy\testing\utils.py", line 665, in assert_array_compare raise AssertionError(msg) AssertionError: Not equal to tolerance rtol=5e-10, atol=0 (mismatch 100.0%) x: array([[ -2.965667e+01, -1.988865e+02, -2.370194e+00, -1.003654e+01], [ -1.988865e+02, -1.383377e+03, -1.592292e+01, -6.800266e+01], [ -2.370194e+00, -1.592292e+01, -8.301699e-01, -8.301699e-01], [ -1.003654e+01, -6.800266e+01, -8.301699e-01, -3.449885e+00]]) y: array([[ -2.965667e+01, -1.988865e+02, -2.370194e+00, -1.003654e+01], [ -1.988865e+02, -1.383377e+03, -1.592292e+01, -6.800266e+01], [ -2.370194e+00, -1.592292e+01, -8.301699e-01, -8.301699e-01], [ -1.003654e+01, -6.800266e+01, -8.301699e-01, -3.449885e+00]]) the suggestion is to add rtol and atol to the mismatch summary, so we can see if it's just a precision issue or something serious rtol = np.max(np.abs(x / y - 1) atol = np.max(np.abs(x - y) (mismatch 100.0% rtol=xxx atol=xxx) (and as aside to the "all close" discussion: I do set the tolerances very carefully especially if the agreement with comparison numbers is below 1e-6 or so) Josef