General procedures are outlined for the simulation and propagation of random and systematic errors in thermophysical property experiments. Density second virial coefficients B(T) from sonic velocity and Joule-Thomson (J-T) experiments are examined for error propagation where the connecting thermodynamic identity is a differential equation with missing boundary conditions. A recent controversy is addressed concerning B(T) at subcritical temperatures for pure hydrocarbon gases from direct density measurements vs. new sonic velocity data. Sonic velocity results are more likely correct with adsorption errors causing the problem in the density measurements. Two new model consistency tests are developed for checking assumed temperature models in the reduction of sonic velocity and J-T data to B(T). Excellent values of B(T) are then obtained from either type of data when the original experiments are free of errors. Random errors propagate systematically when the connecting equation is a differential equation. Sonic data must be of high precision (+/- 10 ppm) to generate B(T) to +/- 1 cm(3)/mol due to complications in data reduction arising from the temperature model/random error interaction. Except perhaps for adsorption errors, systematic errors in the sonic velocities are unimportant to B(T). J-T data provide propagation factors near unity with errors in B(T) higher at higher temperature, unlike sonic velocities.