Seven steel samples previously run by Glow Discharge Spectroscopy were now metallographically investigated by SEM. The aim of the investigation was to see if any features such as cracks, pores or nitrides could be found and correlated to the stability of the GD nitrogen signal. This was indeed the case. The CRM sample JK 2D, previously giving perfect nitrogen stability, had zero measured crack area and the DH 1 (Dillinger Hütte lollipop), being close in stability, only had a very small measured area. All other samples, previously showing scatter above the normal 1 – 1.5 % level, gave appreciably higher crack areas of the order of 0.005%. After the metallographic work the samples were run again spectrometrically to see if the differences in nitrogen scatter between the samples previously established were still the same. This could be verified although the precision ratio for the unstable to stable samples was now reduced from around 4:1 previously to 2:1. It appeared that the special metallographic oxide polishing technique employed for the present investigation caused this reduction although this remains to be verified by continued work. Quite other causes for instabilities and problems were also discussed. It could be shown that certain instabilities on otherwise stable samples were accompanied by high hydrogen signals and it was suggested that the so called H-effect (i.e. the influence of hydrogen on the plasma atomic emission) caused this. It was also shown by calculations how a too limited calibration interval (N ≤ 150 ppm in low-alloyed steel) will amplify the uncertainty of the calibration constants to reach levels far above that of the intensity-scatter of the samples defining them.