Information obtained with Very Long Baseline Interferometry (VLBI) about radio spectra of parsec-scale jets and their evolution can be crucial for distinguishing between various jet models. However, there are several aspects of VLBI which impede spectral studies of parsec-scale regions. The reliability of spectral information extracted from VLBI data depends on many factors including sampling functions (uv-coverages) at different frequencies, alignment of the images, calibration and self-calibration errors, a narrow range of observing frequencies, and source variability. The influences of all these factors must be understood and, if possible, corrected for, in order to reconstruct the spectral properties of parsec-scale jets consistently.

Many of the usual technical problems in making spectral index maps from VLBI data can be avoided by using the VLBA (see Zensus, Diamond, \& Napier 1995). With the VLBA, it is possible to achieve array homogeneity, have an improved flux calibration, and make the time separation between observations at different frequencies negligible. The major remaining problems are image alignment and uneven spatial samplings of VLBI data taken at different frequencies.

To overcome, or at least reduce, the negative effect of uneven uv-coverages, the following scheme of observation and data reduction can be used:

1) Quasi-simultaneous multi-frequency observations.

2) Careful choice of wavelengths. The choice of wavelengths must be a compromise between the possibility of detecting the source structure and the VLBA sensitivity. Typically, at frequencies higher than 22GHz, the requirements on brightness temperature limit the sensitivity. Also, there is a stronger dependence of the high frequency data on atmospheric instabilities. At frequencies lower than 2.3\,GHz, many sources will become too complicated to warrant a successful structure detection without a full--track observation.

3) Applying the phase-cal information for aligning the relative phases in separate frequency bands (Cotton 1995).

4) Improved amplitude calibration, due to frequent system temperature measurements and a relatively weak elevation dependence of the power gains of the VLBA antennas (Moran \& Dhawan 1995).

5) Applying appropriate uv-tapering, in order to provide matching uv-ranges for data taken at different frequencies.

6) Convolving data at different frequencies with the same, artificially circular beam.

7) Using SNR and flux threshold cutoffs, in order to leave out the remaining sidelobe--induced artifacts.

8) If strong sidelobe effects remain present, matching the {\em uv}--coverages within certain uv-ranges or over the whole uv-plane.