This article studies the finite-time formation control problem for networked nonholonomic mobile robots (NMRs) under denial-of-service (DoS) attacks. The communication signals are transmitted among topology-connected NMRs by threatened networks, which might be disconnected or recovered at any time due to DoS attacks. Under DoS attacks, the received discontinuous date packets resulting from the directed graph might make the formation error escape to an unknown boundary. In addition, it is demonstrated that the frequency of DoS attacks directly affects the convergence time of controllers. A novel distributed observer is developed for each follower to narrow the unknown boundary, in which a projection operation is used to restrain the estimation of state with a known bounded set. An event-triggered mechanism is developed for each follower to identify the intermittent DoS attacks by verifying the time stamps of the received data packets. To reduce the effects of the attacker on the closed-loop system, a finite-time formation controller is proposed to guarantee formation convergence to the desired position in every duration of DoS attack-shut, preventing formation error accumulation. The Lyapunov tools are utilized to derive the conditions for guaranteeing the finite-time formation convergence under DoS attacks subject to a certain frequency. It is proved that the state estimation errors and formation errors can be regulated to the neighborhoods of origin in every duration of DoS attacks shut. The DoS attack and distributed DoS attack examples are provided to illustrate the correctness and performance of the proposed method for networked NMRs through numerical simulation comparisons.