Given a graph $G$ and an $r$-edge-colouring $χ$ on $E(G)$, a Hamilton cycle $H\subset G$ is said to have $t$ colour-bias if $H$ contains $n/r+t$ edges of the same colour in $χ$. Freschi, Hyde, Lada and Treglown showed every $r$-coloured graph $G$ on $n$ vertices with $δ(G)\geq(r+1)n/2r+t$ contains a Hamilton cycle $H$ with $Ω(t)$ colour-bias, generalizing a result of Balogh, Csaba, Jing and Pluhár. In 2022, Gishboliner, Krivelevich and Michaeli proved that the random graph $G(n,m)$ with $m\geq(1/2+\varepsilon)n\log n$ typically admits an $Ω(n)$ colour biased Hamilton cycle in any $r$-colouring. In this paper, we investigate colour-biased Hamilton cycles in randomly perturbed graphs. We show that for every $α>0$, adding $m=O(n)$ random edges to a graph $G_α$ with $δ(G_α)\geq αn$ typically ensures a Hamilton cycle with $Ω(n)$ colour bias in any $r$-colouring of $G_α\cup G(n,m)$. Conversely, for certain $G_α$, reducing the number of random edges to $m=o(n)$ may eliminate all colour biased Hamilton cycles of $G(n,m)\cup G$ in a certain colouring. In contrast, at the critical endpoint $α=(r+1)/2r$, adding $m$ random edges typically results in a Hamilton cycle with $Ω(m)$ colour-bias for any $1\ll m\leq n$.