Sequence stratigraphy has been applied in a wide range of scales of time and space, from decimeter-thick layers formed within hours to kilometer-thick basin fills formed during hundreds of millions of years. The traditional approach to practice sequence stratigraphy in this wide range of scales is to subdivide the sediment piles into an ordered hierarchy of sequence cycles of different duration and different architecture. An alternative are scale-invariant models with fractal characteristics. Published data confirm two predictions of the ordered-hierarchy model: sequences of very short duration (<1 × 10³ years) are parasequences bounded by flooding surfaces, very long sequences (>200 × 10⁶ years) are symmetrical transgressive–regressive cycles. However, the sequence record in the range of 1 × 10⁴–200 × 10⁶ years, the principal domain of sequence stratigraphy, shows a rather irregular succession of sequences with variable symmetry and bounded by flooding surfaces or exposure surfaces. For these time scales, scale-invariant models are a good first approximation, particularly because the evidence for scale-invariance and randomness in the stratigraphic record is strong: Frequency spectra of sea-level change as well as rates of sedimentation and rates of accommodation change plotted against length of observation span show basic trends indistinguishable from random walk. These trends, combined with scale-invariant sequence models may be the most efficient tools for across-the-board predictions on sequences and for locating islands of order in the sequence record.