The spectra were de-redshifted assuming *z* = 0.0434 and analyzed in terms
of Gaussian components as described above. We discovered first that the core
of the lines could not be fitted by a single set of narrow Gaussian profiles.
To get a satisfactory fit, two sets of Gaussian components are needed: the
first, unresolved (and subsequently taken as the origin of the velocity
scales) has 6583/H = 0.55,
5007/H = 1.27, and corresponds to a HII region; the second
is resolved (FWHM 350 kms^{-1}, corrected for the instrumental
broadening), blueshifted by 95 kms^{-1} with respect to the
narrow components and has line intensity ratios typical of a Seyfert 2
(6583/H = 0.84, 5007/H > 10).

At this stage, we removed from the blue and red spectra the best fitting core (the HII region and the Seyfert 2 nebulosity, Fig. 1c and d), obtaining two spectra we shall call ``original data - core''. The blue one was then fitted with a broad H Gaussian component and two sets of three components modeling the narrow H and [OIII] lines. The result is very suggestive: one set has a strong H line and very weak negative [OIII] components, while the other set displays a strong [OIII] contribution and a weak negative H component, showing that we have in fact a H component with no associated [OIII] emission and [OIII] lines with a very weak (undetected) associated H; in other words, the region producing the H line does not emit forbidden lines, while the [OIII] emitting region has a high 5007/H ratio, which are the characteristics of the ``broad'' and ``narrow'' line regions in Seyfert 1 galaxies, respectively.

With these results in mind, we optimized this last fit by using a Lorentzian
profile for the H line, with no associated [OIII] emission, and a set
of three Gaussians for the remaining contribution coming from the ``narrow''
components; to avoid an unphysical negative intensity for the H line,
we forced 5007/H to be equal to 10, which is the ratio
usually found for the narrow component in Seyfert galaxies. The best fit is
presented in Fig. 1e. The H Lorentzian component is blueshifted by
160 kms^{-1} with a width of 920 kms^{-1}; the [OIII] lines are
blueshifted by 395 kms^{-1} and their width is 1120
kms^{-1}.

We have also analyzed the ``original data - core'' red spectrum (Fig. 1f)
with one Lorentzian H component and a set of three Gaussians (for the
H and [NII] lines) with the constraint that
6583/H = 0.9, for which we have found a FWHM of
770 kms^{-1} and a blueshift of 375 kms^{-1}. The H Lorentzian
component, blueshifted by 55 kms^{-1}, has a width of 1030
kms^{-1}, in reasonable agreement with the width of the corresponding
H component. The Lorentzian Balmer components, without any measurable
associated forbidden line, would qualify KUG 1031+398 as a NLS1 with, in fact,
very narrow lines. The other system of lines, with a very high
5007/H ratio, 6583/H 0.9 and
FWHM 945 kms^{-1}, is analogous to what is usually found in
Seyfert 2s and corresponds to a NLR cloud.

At last, we fitted the ``original data - core'' blue spectrum with a broad H Gaussian component and one set of three Gaussians (modeling H and the [OIII] lines) for which we set the 5007/H ratio to the value found by Mason et al. for the ``intermediate'' component, i.e., 1.42. The red spectrum was fitted with two H components, for which we fixed the redshifts to the values obtained in the blue spectrum profile fitting analysis. The resulting fits and residuals, shown in Figs. 1g and h, seem to be significantly worse than the ones given in Figs. 1e and f, showing that the presence of an ``intermediate'' component is not required by the data.