Here we investigated how well internal estimates of direction of gravity are preserved over time and if the subjective visual vertical (SVV) and horizontal (SVH) can be used inter-changeably. Fourteen human subjects repetitively aligned a luminous line to SVV, SVH or subjective visual oblique (± 45°) over 5 min in otherwise complete darkness and also in dim light. Both accuracy (i.e., the degree of veracity as reflected by the median adjustment error) and precision (i.e., the degree of reproducability as reflected by the trial-to-trial variability) of adjustments along the principle axes were significantly higher than along the oblique axes. Orthogonality was only preserved in a minority of subjects. Adjustments were significantly different between SVV vs. SVH (7/14 subjects) and between ±45° vs. -45° (12/14) in darkness and in 6/14 and 14/14 subjects, respectively, in dim light. In darkness, significant drifts over 5min were observed in a majority of trials (33/56). Both accuracy and precision were higher if more time was taken to make the adjustment. These results introduce important caveats when interpreting studies related to graviception. The test re-test reliability of SVV and SVH can be influenced by drift of the internal estimate of gravity. Based on spectral density analysis we found a noise pattern consistent with 1/fβ noise, indicating that at least part of the trial-to-trial dynamics observed in our experiments is due to the dependence of the serial adjustments over time. Furthermore, using results from the SVV and SVH inter-changeably may be misleading as many subjects do not show orthogonality. The poor fidelity of perceived ± 45° indicates that the brain has limited ability to estimate oblique angles.