Current global fits to the body of DIS and other data used to extract the parton distribution functions of the nucleon do not propagate experimental errors, and thus do not yield error estimates on the PDFs or on subsequent theoretical predictions using those PDFs. I discuss a general method that allows taking the complete experiment errors into account when extracting PDFs, and that produces estimates of errors in either fit PDF parameters or in subsequent theoretical predictions of collider observables. I will also present an initial numerical analysis using the method.