The design of robust countermeasures against Side-Channel Analysis or Fault Attacks is always a challenging task. At WISTP’14, a single countermeasure designed to thwart in the same effort both kinds of attacks was presented. This countermeasure is based on coding theory and consists in a specific encoding of the manipulated data acting in the same time as a random masking and an error detector. In this paper, we prove that this countermeasure does not meet the ambitious objectives claimed by its authors. Indeed, we exhibit a bias in the distribution of the masked values that can be exploited to retrieve the sensitive data from the observed side-channel leakage. Going further, we show that this bias is inherent to the nature of the encoding and that randomizing the code itself can be useful to reduce the bias but cannot completely fix the scheme.