Abstract: Complex dynamical systems often show sudden major changes, or tipping points, as the system gradually changes. Examples include mass extinctions in an ecosystem, deforestation, and aggressive progression of a disease in a human body. Exploiting critical slowing down phenomena among other things, various early warning signals (EWSs) that anticipate tipping events before they occur have been developed. In fact, complex dynamical systems of our interest often form a heterogeneous network, often as a result of emergent behavior of interacting agents. How to construct EWSs in this network situation is not straightforward. We present heuristic and mathematically grounded methods to select sentinel nodes in a given network to construct good EWSs. We show that carefully chosen small subsets of nodes can anticipate transitions as well as or even better than using all the nodes under a wide variety of conditions. We further address the case in which one cannot use too many observations from each node for forming EWSs.