I slightly adapted the tutorial for TLinearFitter, so that it would read data from a file. Unfortunately, this seems to lead to problems. The macro below boils down the problem. If one runs the without arguments, it generates random data, writes data to a file (data.txt), and then fits it fine. If you then run it with as test(kFALSE), it will read the data from the file and fit it - and fail. 
There is a commented section that dumps the data points to the screen to verify they are the same in both cases.
Cheers,
Chris
#include <TRandom.h>
#include <TLinearFitter.h>
#include
#include
#include
using namespace std;
void test( Bool_t random=kTRUE ) {
const Int_t n=100;
Double_t x[n102];
Double_t y[n10];
Double_t e[n10];
if ( random ) {
TRandom rand;
ofstream outfile(“data.txt”);
for (Int_t i=0; i<n; i++){
x[0 + i2] = rand.Uniform(-10, 10);
x[1 + i2] = rand.Uniform(-10, 10);
e[i] = 0.01;
y[i] = 4x[0+i2] + x[1+i2] + rand.Gaus()e[i];
outfile << setw(20) << x[0+i2]
<< setw(20) << x[1+i2]
<< setw(20) << y[i] << endl;
}
}
else {
ifstream infile(“data.txt”);
for (Int_t i=0; i<n; i++){
infile >> x[0+i2]
>> x[1+i2]
>> y[i];
infile.ignore(8192,’\n’);
}
}
// for (Int_t i=0; i<n; i++){
// cout << setw(20) << x[0+i2]
// << setw(20) << x[1+i2]
// << setw(20) << y[i] << endl;
// }
TLinearFitter lf(2);
lf.SetFormula(“1 ++ x0 ++ x1 ++ x0x0 ++ x0x1 ++ x1*x1”);
lf.AssignData(n, 2, x, y, e);
lf.Eval();
for (Int_t i=0; i<lf.GetNumberFreeParameters(); i++)
printf(“par[%d]=%f±%f\n”, i, lf.GetParameter(i),
lf.GetParError(i));
Double_t chisquare=lf.GetChisquare();
printf(“chisquare=%f\n”, chisquare);
}