We present a detailed analysis of the horizontal branch of the Carina dwarf spheroidal galaxy by means of synthetic modelling techniques, taking consistently into account the star formation history and metallicity evolution as determined from main sequence and red giant branch spectroscopic observations. We found that a range of integrated red giant branch mass loss values of 0.1-0.14 M increasing with metallicity is able to reproduce the colour extension of the old horizontal branch. Nonetheless, leaving the mass loss as the only free parameter is not enough to match the detailed morphology of Carina horizontal branch. We then investigated the role played by the star formation history on the discrepancies between synthetic and observed horizontal branches. We derived a "toy" bursty star formation history that reproduces well the observed horizontal branch star counts, and also matches qualitatively the red giant and the turn-off regions. This bursty star formation history is made of a subset of age and [M/H] components of the star formation history based on turn off and red giants only, and entails four separate bursts of star formation of different strengths, centred at 2, 5, 8.6, and 11.5 Gyr, respectively, with mean [M/H] decreasing from ∼−1.7 to ∼−2.2 when the age of the burst increases, and with a Gaussian spread of σ 0.1 dex around these mean values. The comparison between the metallicity distribution function of our bursty star formation history and the one measured from the infrared CaT feature using a CaT-[Fe/H] calibration shows a qualitative agreement, once the range of [Ca/Fe] abundances measured in a sample of Carina stars have been taken into account, that causes a bias of the derived [Fe/H] distribution toward values that are too low. In conclusion, we show how the information contained within the horizontal branch of Carina (and dwarf galaxies in general) can be extracted and interpreted to refine the star formation history derived exclusively from red giants and turn-off stars.