I find that the following script consistently core dumps (scipy latest CVS, Numeric 23.7, ATLAS 3.6.0) on both FC1 and FC3 when I say "python script.py" at the command line. I was trying to test problems others have reported using the Jacobian argument to leastsq, including a message way back in April 2003 from John Hunter. I'm pretty familiar with MINPACK, having used it to fit some functions a while back. Where might I start looking for the problem? from Numeric import * import scipy as S def ros(x): return array([10*(x[0] - x[1]**2), 1-x[1]]) def dros(x): return array([[-20*x[0], 10],[-1,0]]) def dros_trans(x): return array([[-20*x[0], -1],[10,0]]) x0=[-1.2,1.0] print "No Jacobian:" print S.optimize.leastsq(ros,x0) print "With Jacobian, try 1:" print S.optimize.leastsq(ros,x0,Dfun=dros) print "With Jacobian, try 2:" print S.optimize.leastsq(ros,x0,Dfun=dros_trans) print "With Jacobian, try 3:" print S.optimize.leastsq(ros,x0,Dfun=dros,col_deriv=1) print "With Jacobian, try 4:" print S.optimize.leastsq(ros,x0,Dfun=dros_trans,col_deriv=1)