We aim to investigate the propriety of stellar parameter errors of the official data re- lease of the LAMOST low-resolution spectroscopy (LRS) survey. We diagnose the errors of radial velocity (RV), atmospheric parameters ([Fe/H], T eff , log g) and α-enhancement ([α/M]) for the latest data release version of DR7, including 6,079,235 effective spectra of 4,546,803 stars. Based on the duplicate observational sample and comparing the deviation of multiple measurements to their given errors, we find that, in general, the error of [α/M] is largely un- derestimated, and the error of radial velocity is slightly overestimated. We define a correction factor k to quantify these misestimations and correct the errors to be expressed as proper inter- nal uncertainties. Using this self-calibration technique, we find that the k-factors significantly vary with the stellar spectral types and the spectral signal-to-noise ratio (SNR). Particularly, we reveal a strange but evident trend between k-factors and error themselves for all five stellar parameters. Larger errors tend to have smaller k-factor values, i.e., they were more overesti- mated. After the correction, we recreate and quantify the tight correlations between SNR and errors, for all five parameters, while these correlations have dependence on spectral types. It also suggests that the parameter errors from each spectrum should be corrected individually. Finally, we provide the error correction factors of each derived parameter of each spectrum for the entire LAMOST-LRS DR7.